Automotive Chip Shortage is as much about the "type of chips" as it is about overall capacity

[JForce]

 

Summary

Chip manufacturers are pointing out (and working with) to the automotive industry that they need to update the design of the electronics they use in their cars, as setting up to manufacture higher volumes of old designs simply isn't worth it

 

Quotes

Quote

 “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.” 

 

My thoughts

Even as someone who messes around with cars and technology, it's surprising to me how obvious this is now that it's been pointed out.  Car manufacturers are the epitome of "if it ain't broke don't fix it" as far as production costs go. Add to that the very high bar they require for resiliency and reliability, and it's understandable that once they've got a working design for a given control component, there's little incentive for them to iterate on it.  However there's only so much capacity for manufacturing silicon around the world, and with so much consumer electronic demand, foundries and associated companies wouldn't see much point in investing to build old stuff.

 

We've talked about the silicon shortage and the millions of vehicles sitting around in factories, finished except for control modules waiting on components, but it does seem like we missed (or at least I know I did) the fact that the issue is overall capacity rather than specific capacity.

 

As the article states, Intel/AMD etc will keep pumping out as many "modern"  chips as you want, as they go into everything.....but to increase capacity for old stuff makes no sense. <Insert joke about Intel using old huge designs anyway>

 

Sources

https://fortune.com/2021/09/17/chip-makers-carmakers-time-get-out-semiconductor-stone-age/

[Zodiark1593]

I kind of disagree. Reliability really trumps just about everything else when it comes to vehicles, especially as a number of functions are being controlled by the infotainment system. 
 

If vehicle makers can move to newer processes in a timely manner without compromising reliability (which will take time and tons of validation), then by all means do so. However, rushing over to new silicon is not the way to go. And with drive-by-wire slowly coming into play, failures are really not an option. 

[The_russian]

17 minutes ago, Zodiark1593 said:

And with drive-by-wire slowly coming into play, failures are really not an option. 

Agree with your post, but did you mean brake-by-wire? Drive-by-wire has been around for some time and failure is even more not an option for brake-by-wire. 

[Zodiark1593]

52 minutes ago, The_russian said:

Agree with your post, but did you mean brake-by-wire? Drive-by-wire has been around for some time and failure is even more not an option for brake-by-wire. 

Throttle-by-wire has been a thing for awhile, though I also refer to steering as well. 

[leadeater]

1 hour ago, Zodiark1593 said:

Throttle-by-wire has been a thing for awhile, though I also refer to steering as well. 

It applies to more and more as well; Lane keep assists, ESD, ABS, TC, Radar Cruise and Radar Breaking (happening when not supposed to and the reverse of course).

 

A lot of functions are moving over to digital sensors and computer systems in cars, faults and bugs and how they are handled is really important.

[BlueChinchillaEatingDorito]

2 hours ago, Zodiark1593 said:

Throttle-by-wire has been a thing for awhile, though I also refer to steering as well. 

Basically every thing in terms of the drivetrain of a vehicle is "by-wire" now. The only thing I can think of that might not be is your parking brake (but that's also slowly being phased out even on commuter vehicles) and steering if it's old school rack and pinion and maybe hydraulic. 

[Kisai]

35 minutes ago, BlueChinchillaEatingDorito said:

Basically every thing in terms of the drivetrain of a vehicle is "by-wire" now. The only thing I can think of that might not be is your parking brake (but that's also slowly being phased out even on commuter vehicles) and steering if it's old school rack and pinion and maybe hydraulic. 

Power Steering has been on ALL vehicles since the 90's, It's simply impossible to drive a vehicle that doesn't have it because the amount of muscle strength needed to turn at low speeds is too high for most people who didn't grow up on it. Even then, that's not an IC thing, that's a hydraulic system. (try steering a car at 5-30kph, and then try the same thing with a riding lawntractor or forklift, typically the steering is MUCH harder than a car.)

 

I don't think people realize that all vehicles that have a steering wheel are a mixture of hydraulic "by wire" systems, not electronic (like in an air plane.) Even from a safety POV, never mind cost, you still want to be able to pull the parking brake, put the transmission in Park/Neutral, and Steer even if all the electronics fail, otherwise you won't be able to tow it without destroying the entire drivetrain (which is why tow trucks lift the front/rear of a vehicle instead of simply dragging it.) 

 

Now with EV's, like not hybrids, but full EV's, you have the potential option of being entirely electric just to be more reliable than a typical ICE vehicle, but vehicles that use this (Tesla's) have multiple redundancies. I'm not sure if I would trust a "cheap" vehicle that is entirely by wire with no redundancy, but then I remember that the Skytrain has been a fully electric, fully autonomous system since the 1980's and it's never had an accident attributable to the automation, where as cheaper at-grade human-driven light-rail systems tend to have daily accidents ranging from over-shooting the stop to dragging people to their death caught in the doors.

 

Then you have the hacking problem. Yes, all computer driven systems can be hacked, it's not a question of IF, but when/where. Using the previous Skytrain example, despite it using rather off-the-shelf RF technology, has never been "hacked". Where as again, human drivers have managed to speed and over-apply brakes, overshooting stops and causing excessive wear on the braking systems. (When the Skytrain is manually driven during extreme cold weather events, it's a completely different over-cautious low-speed experience.) Automated vehicles are more likely to be vandalized than hacked.

 

So I think we can trust automation for driving, we can trust drive-by-wire, but we need to have a long talk with vehicle manufacturers to standardize on ONE, and only ONE standard that takes into redundancy of both the city's traffic control system and onboard systems as well as talks to neighboring cars to make sure it gets as much advanced notice as possible before passing vehicles that are turning and such, and I feel it's this is where vehicle manufacturers are going to drop the ball and rely too much on optical machine learning rather than just having the car go "I need to turn in 100m, watch out."

 

[Tenelia]

Just for learning's sake: Could someone shed light on how long it'd take to create similar resilient and robust designs on a 5nm node?

[RedRound2]

Is it really that difficult to migrate to new chips and validate it? Plus modern processor will also give a huge boost in performance and responsiveness to infotainment system which has historically been so bad in almost all cars except Teslas.

 

I figure it's more about laziness than anything. Testing out ECUs for a long period of time shouldn't be such a multi year time consuming process. And having redundancies will also help to huge degree

[KWelz]

29 minutes ago, Caroline said:

Cars with chips are a nightmare, imagine a software update or a firmware lock preventing you from firing up the engine.

Cars have used microprocessors for almost 40 years.  

[leadeater]

1 hour ago, Kisai said:

I don't think people realize that all vehicles that have a steering wheel are a mixture of hydraulic "by wire" systems, not electronic (like in an air plane.)

All current power steering is electric, last hold outs like Porsche moved from hydraulic to electric quite a few years ago now. Only car platforms that have not been updated in the last 8 years run hydraulic power steering, electric is much cheaper and lighter and smaller.

[CarlBar]

17 minutes ago, RedRound2 said:

Is it really that difficult to migrate to new chips and validate it? Plus modern processor will also give a huge boost in performance and responsiveness to infotainment system which has historically been so bad in almost all cars except Teslas.

 

I figure it's more about laziness than anything. Testing out ECUs for a long period of time shouldn't be such a multi year time consuming process. And having redundancies will also help to huge degree

 

The problem is the only way to validate it will last and continue to function despite extreme conditions is to exposes it to those extreme condition for the appropriate length of time. You can shortcut it somewhat if the item isn't used 24 hours a day, (as most vehicles aren't). But for cars if you want to cover reasonable heavy daily driving you probably need to cover the equivalent of 6 hours a day IMO. That means for every 4 years of validation you have a minimum of a years testing time required in a dedicated, (and thus expensive), automated test environment.

[Kadzo]

2 hours ago, BlueChinchillaEatingDorito said:

Basically every thing in terms of the drivetrain of a vehicle is "by-wire" now. The only thing I can think of that might not be is your parking brake (but that's also slowly being phased out even on commuter vehicles) and steering if it's old school rack and pinion and maybe hydraulic. 

Parking locks are moving to electric, i have worked on a project for my previous company for those e brakes. Basically its a servomotor on all 4 wheels and it receives info from the car system to activate the lock.

[RedRound2]

44 minutes ago, CarlBar said:

 

The problem is the only way to validate it will last and continue to function despite extreme conditions is to exposes it to those extreme condition for the appropriate length of time. You can shortcut it somewhat if the item isn't used 24 hours a day, (as most vehicles aren't). But for cars if you want to cover reasonable heavy daily driving you probably need to cover the equivalent of 6 hours a day IMO. That means for every 4 years of validation you have a minimum of a years testing time required in a dedicated, (and thus expensive), automated test environment.

4 years of validation seems a bit much. I doubt anything on the street today was validated for four years before it made to production. Is there any source for this? All these car manufactureres today could make a 100 of these vehicles in probably the second most advanced cutting edge tech (7nm) and test for a year or so. If there's issues, test more, otherwise you slowly start transitioning it. Seems like a very simple solution to me.

 

And what exactly is expected to fail again by just a simple node swap? The environements we're talking about is regular earth atmosphere and not space or something like that where radiations could do a lot of damage on a smaller node processor

[Kisai]

58 minutes ago, RedRound2 said:

4 years of validation seems a bit much. I doubt anything on the street today was validated for four years before it made to production. Is there any source for this? All these car manufactureres today could make a 100 of these vehicles in probably the second most advanced cutting edge tech (7nm) and test for a year or so. If there's issues, test more, otherwise you slowly start transitioning it. Seems like a very simple solution to me.

 

And what exactly is expected to fail again by just a simple node swap? The environements we're talking about is regular earth atmosphere and not space or something like that where radiations could do a lot of damage on a smaller node processor

I'm not sure if you're aware, but electronics literately come in two varieties, one of them being automotive.

 

https://www.intel.com/content/dam/www/programmable/us/en/pdfs/literature/pt/cyclone-v-product-table.pdf

 

The Cyclone V is 28nm from 2012 btw. Basically think Core i3-3x00 size.

 

 

https://www.intel.ca/content/dam/www/programmable/us/en/pdfs/literature/hb/auto/automotive_handbook.pdf

on page 4:

Quote

Intel provides certified Automotive Functional Safety Data Package (AFSDP) for devices that comply with ISO 26262. AFSDP delivers the framework, methodology, tools, and IP to assist you in building a safe system with cost and time savings. AFSDP typically saves you 12-18 man-months in certifying your safety critical applications at system level.

 

Automotive CLPD/FPGA's have a spec of "TJ = -40°C to 125°C" where as non-automotive models are less.

[Levent]

1 hour ago, CarlBar said:

That means for every 4 years of validation you have a minimum of a years testing time required in a dedicated

If they are just going to move their already existing BCM designs to newer nodes while having the new parts as drop in replacements, this really shouldnt take that long. All manufacturers test multiple vehicles at once to gather combined test time, not individual test time.

[CarlBar]

1 hour ago, Levent said:

If they are just going to move their already existing BCM designs to newer nodes while having the new parts as drop in replacements, this really shouldnt take that long. All manufacturers test multiple vehicles at once to gather combined test time, not individual test time.

 

Thats the point, they can't, they're going to have to come up with all new designs and those have to be certified for safety reasons.

 

2 hours ago, RedRound2 said:

4 years of validation seems a bit much. I doubt anything on the street today was validated for four years before it made to production. Is there any source for this? All these car manufactureres today could make a 100 of these vehicles in probably the second most advanced cutting edge tech (7nm) and test for a year or so. If there's issues, test more, otherwise you slowly start transitioning it. Seems like a very simple solution to me.

 

And what exactly is expected to fail again by just a simple node swap? The environements we're talking about is regular earth atmosphere and not space or something like that where radiations could do a lot of damage on a smaller node processor

 

Many cars are offered warranties that long. And if safety critical things start failing that fast regulatory agencies will start taking notice and voicing their displeasure.

 

Your home computer doesn't have to deal with vibration, shock loads, (pothole's, e.t.c.), extreme's of temperature, contact with dirt grime an corrosive stuff, (salt is a particular concern in colder climates), not to mention potential engine induced RF noise and a much less clean power feed than your average mains power system. All with little or no maintenance for years on end.

[Tenelia]

24 minutes ago, James Evens said:

One of the reason why they don't use cutting edge is cost. 

You have more or less a curve where the cutting edge is extremely expensive and very old ones are too getting costly. The sweet spot was somewhere around 28nm.

What're the nodes currently manufactured actively for automakers?

[Zodiark1593]

6 hours ago, RedRound2 said:

Is it really that difficult to migrate to new chips and validate it? Plus modern processor will also give a huge boost in performance and responsiveness to infotainment system which has historically been so bad in almost all cars except Teslas.

 

I figure it's more about laziness than anything. Testing out ECUs for a long period of time shouldn't be such a multi year time consuming process. And having redundancies will also help to huge degree

Porting designs over to a new process is seldom a straightforward process, as different manufacturing techniques impose differing constraints and design rules upon chip architects. Alterations to the architecture would be required during the porting process. 
 

This is where potential for failure can be introduced. 

[Donut417]

Maybe automakers need to invest in their own Fabs and make the chips themselves if the current chip makers dont see it as being practical. 

[CarlBar]

4 hours ago, Tenelia said:

What're the nodes currently manufactured actively for automakers?

 

According to the article in the OP 45-90nm. For sufferance the last models of Pentium 4 and the first multi-core Intel CPU's where on 65nm.

[StDragon]

You need chips that are hardened for their environment. It's my limited understanding that if your chip is 45nm, then you need to redo the design and mask for say 5nm. However, the gates don't need to be that small. The node is basically optical resolution; not that you have to make the transistor as small as possible.

[RedRound2]

7 hours ago, Kisai said:

I'm not sure if you're aware, but electronics literately come in two varieties, one of them being automotive.

 

https://www.intel.com/content/dam/www/programmable/us/en/pdfs/literature/pt/cyclone-v-product-table.pdf

 

The Cyclone V is 28nm from 2012 btw. Basically think Core i3-3x00 size.

 

https://www.intel.ca/content/dam/www/programmable/us/en/pdfs/literature/hb/auto/automotive_handbook.pdf

on page 4:

 

Automotive CLPD/FPGA's have a spec of "TJ = -40°C to 125°C" where as non-automotive models are less.

 

I am aware of the different standards and diffierent minimum requirements for automotive, military or space. But is it mentioned anywhere that any new tech needs to be tested for a minimum of 4 years before it hits the road? That's what I wanted to know. I get the temperature, durability, reliability aspect of the certifications. That I assume you can also get or request to get with 7nm or 10nm processors today from Intel, ARM and Nvidia -  atleast as per this news

2 hours ago, Zodiark1593 said:

Porting designs over to a new process is seldom a straightforward process, as different manufacturing techniques impose differing constraints and design rules upon chip architects. Alterations to the architecture would be required during the porting process. 
 

This is where potential for failure can be introduced. 

Yeah, i understand the architecture needs to be changed as well most of the time.

 

But ECUs are failry straighforward devices with straightforward functionalities. It's actualy companies like Teslas Im more worried about this since their computer pretty much handles almost all aspects of the car. But other cars usually have a infortainment center (that controls the AC at max) and various ECUs for cruise control and any other lower level basic functions. Granted mid to high end cars these days have more electronics control, but still not sure how complicated it gets. Definitely not as complicated as self driving which is basically being run on more cutting edge chips (and has more consequences if things go wrong)

[Levent]

30 minutes ago, RedRound2 said:

But ECUs are failry straighforward devices with straightforward functionalities. It's actualy companies like Teslas Im more worried about this since their computer pretty much handles almost all aspects of the car. But other cars usually have a infortainment center (that controls the AC at max) and various ECUs for cruise control and any other lower level basic functions. Granted mid to high end cars these days have more electronics control, but still not sure how complicated it gets. Definitely not as complicated as self driving which is basically being run on more cutting edge chips (and has more consequences if things go wrong)

ECU = Engine Control Unit, which strictly controls the engine.

TCU/TCM = Transmission Control Unit/Module, Only on DCT/Auto/CVT vehicles and only controls transmission behaviour.

BCM = Body Control Module, controls user inputs, reads non engine related sensors (door ajar, AC temperature, air quality etc)

Infotainment system on cars are merely a IO devices, everything is done on BCMs. EVs differ in some ways but they are not that different. You can literally run most 1-2-3-4-5-6 cylinder engines with an arduino (Check out Speeduino if you are interested), as you said ECUs are relatively simple devices, they have coil outputs, fuel injector outputs, ignition maps, fuel maps, boost control for boosted vehicles, sensor inputs, throttle input and throttle body output (in case of DBW) and a canbus. Cruise control on every combustion engine car I know of are controlled via BCM, not the ECU or the infotainment, as I said ECU strictly controls engine, nothing else. Information is carried via CAN to the BCMs, then to ECU.

[StDragon]

42 minutes ago, RedRound2 said:

But ECUs are failry straighforward devices with straightforward functionalities. It's actualy companies like Teslas Im more worried about this since their computer pretty much handles almost all aspects of the car. But other cars usually have a infortainment center (that controls the AC at max) and various ECUs for cruise control and any other lower level basic functions. Granted mid to high end cars these days have more electronics control, but still not sure how complicated it gets. Definitely not as complicated as self driving which is basically being run on more cutting edge chips (and has more consequences if things go wrong)

It's not just the ECU, but controllers in the transmission, breaking system, sensors, etc. All of which interconnect along the CAN bus. So when talking about the "automotive industry", the real question is which supply chain/s are being effected here?