Intel in talks to produce chips for automakers within six to nine months
reuters.com
reuters.com
I was, many years ago, employed as a manufacturing engineer in semiconductor fabs that, once no longer cutting edge enough to do the latest CPU's or memory chips, would transition to doing higher-voltage stuff. It's a reasonable thing to do. It does not necessarily speak well of Intel's current situation, though, with respect to CPU and memory chips, that they consider this a good move. But, given their current situation, it probably is making the best of a bad situation.
https://www.statesman.com/story/business/2021/03/30/samsung-...
Fabs use a lot of water and electricity. Samsung announced that it was considering a new investment in Austin, then the Texas storms hit. A few days later, Samsung announced that it was considering a new investment in Austin or somewhere else.
Intel already operates several fabs in the US and other first world countries so I don't know what onshore would mean in this context, but Gelsinger has made it pretty clear that he believes our best bet is to make chips at American fabs run by American companies. So I guess that is a yes.
This is not necessarily something you are just good at because you already produce semis.
Auto manufacturers need 380nm production lines.
They unwisely canceled all their orders for >6 months.
Their suppliers mothballed all these antiquated facilities.
Maybe Intel can make money with this. It will be interesting to see.
These guys hate inventory with a raging passion.
Efficiency and Robustness are very often a tradeoff, the more efficient you are, the more prone your entire operation is to minor disruptions.
Something akin to overfitting in machine learning.
Also something to consider is what happens when the entire economy is forced in that direction, where you either cut everything to the bone for the sake of efficiency, or go out of business due to competition.
The longer the lag between producing a faulty part and trying to use it, the harder to diagnose the problem.
The analogous software problem is probably having database backups sitting around without regularly testing if you can successfully restore from them.
And I really wished I the "everything is running on JIT" meme to die. Because it is just plain wrong.
That sounds like a contractor milking a cost-plus contract.
I don't know why someone would agree to them, but they're fairly common. Nasa awarded Boeing a cost-plus contract for the SLS.
Basically they're awarded a contract for the cost estimate plus any overruns, it's basically a blank check.
The problems are, that they stock to much of some parts and not enough from others. That problems gets worse when programs reach end of life, obsolescence is a severe problem then.
Most of the demand fluctuations in aerospace, less so for spares, is home made. Delivery lead times and production lead times are so long, that done right, fluctuations are mainly a problem for long lead time items. And that is manageable.
Overall, each of the projects had around 2000 unique parts (at least for what this particular plant was making). Occasionally, newer contracts would get a much higher prioritization and could grab parts (and employees) from older projects. Another scenario that would lead to extra demand was having test failures on a project that was nearing design freeze.
I ended up changing industries to financial services, but will always have a deep respect and appreciation for what went into those projects.
Overall I felt like there were a number of good strategies but that sometimes the strategizing interfered with the execution. Peoples tenure and the projects they went through often influenced their notion of ideal. By the time I started getting comfortable understanding these things, it was time to get a new job.
Is there a reason for that? I would imagine they'd have actuarials telling them exactly how much of each thing they need to produce a thing, as well as failure rates.
In the end, it is the tram and person responsible for inventory planning and placing production / purchase orders that set, directly and indirectly, inventory levels. In some cases 2 months maybe just right. In other cases it maybe 1 month to much, in again other cases not close to enough. And that can change over time.
As a rule of thumb, you don't want inventory. It makes an operation inflexible and fat. You need it so. Figuring that out is the task planners and supply chain managers. Some orgs are better at that then others.
They "need" chips made on a 28 year old process? I'd really want a source for this. Even if it were true, there are still fabs that can make .35um process chips.
RAD6000 CPUs are still made on 0.5um fabs.
Whos to say the full process to swap out a handful of chips within the car and getting any required tests and certifications done wouldn't take as long or longer than waiting for the next production run of the spec'd chips?
I think you mean 350nm.
This isn't surprising; a lot of nifty features got axed after 350nm, like having two full-fledged polysilicon layers (a few subsequent processes technically have two poly layers but one of them is totally crippled and can only be used for flash cells).
The suppliers didn't mothball the facilities (which are not at all antiquated, and are massively in use for other things), they just sold that capacity to less capricious customers.
LOL. Any code that relies on fast interrupts is a lost cause on x86. A 32 MHz ECU may dramatically faster at interrupts that a top of the line Xeon :)
The power consumption of a pile of Xeons will add up fast, too.
The military has been using automotive grade parts to reduce costs where possible.
https://www.dla.mil/Portals/104/Documents/LandAndMaritime/V/...
I couldn't resist. The joke was just too easy.
They also announced [1] that they were building two new fabs in Arizona a couple weeks ago. There may just be growing demand for shortened supply chain in the semiconductor space (recent trade/tariff wars, chip shortages from offshore fabs, various national security implications)
[1] https://www.forbes.com/sites/tiriasresearch/2021/03/23/intel...
Also, if they want to get busy with foundry, this is the way to do it. It's not like they're doing HPC processors with all the trailing node fabs.
I know an engineering researcher who uses a shaft position sensor chip that is also widely employed in EVs. They have been out of stock for months. That little chip could have major implications for many industries and communities.
"For want of a nail..."
I think that Intel does have two plays left.
1. Appeal to the government that they can't go beyond their current process node and the government preorders a bunch of processors to do this.
2. Intel abandons new process nodes and becomes a customer of TSMC.
If Intel thinks about doing #2 I think the government would be forced to do #1 for the sole fact of keeping chip manufacturing knowledge in the US
But they do need a little bit of option #2, if only for market share defense against competitors. TSMC obviously prioritizes long-term partnership, but in today's demand environment, every wafer that Intel buys is a wafer that AMD (or Nvidia) cannot.
More importantly what do you think China PR thinks the US will do if it marches into Taiwan?
It's a perfect example of cutting one's nose to spite one's face. Also, Taiwan will end up a large, long-term loser of that, as well.
China's ambitions on Taiwan have nearly zero to do with TSMC's presence there. (And everything to do with the past 80 years of history - more specifically, its desire to 'conclude' the civil war. If Taiwan's didn't have a chip industry, the PRC would still have the same goals.)
Sure and the fabs that get spun up will not be under CCP control which is a clear win.
There's no guarantee that's going to be the outcome, and even if it was, it's not one that I'm interested in paying the cost of admission to.
The actual chips are slower because they don't have nearly as many cores, so for workloads that can be highly parallelized they're behind. But that's process.
Delivering Military IC's to a contract is very different to creating commercial products suitable for laptops, phones, AI etc.
> "For want of a nail..."
I remember reading an article about how production of Apple Macintosh desktop computers in Texas stopped for the want of a specific kind of screw. If I remember correctly, in China they would have been able to resume work within hours in the worst case.
I have no industry knowledge so no idea of this is true though.
https://www.nytimes.com/2019/01/28/technology/iphones-apple-...
Tim Cook's Apple is a logistics powerhouse. Surely they have full control over their entire supply chain. Every detail planned long ahead of time.
Which screw part was it? Quick scan of all the popular reports doesn't identify the specific screw.
Why not change the design to use a more readily available screw?
Why not make a one time order of a box of screws from China?
Apple regularly provides the capital so their suppliers can ramp up. How much would have cost for that tiny Austin screwer suppler to double production? Probably less than the loose change found in Cook's office couch.
In conclusion, I don't buy this excuse. I dimly recall another report about the struggles the Austin plant had producing the Mac Pros. Bad fit and finish. Long turn arounds because there's no surrounding ecosystem of skills, suppliers, deep experience. Etc. Nothing that Apple couldn't have fixed; they just didn't want to.
Sounds about right. When my brother was green in his automotive supply logistics career, there were a couple of times he got thrown on an airplane with a carry-on of parts...
Fast forward some years later and I was working at a motion sensor company for consumer electronics. Our node sizes were like 40-50nm and many generations behind cutting edge ICs. Lower margins but meaningful volume. Our fab was naturally, you guessed it, TSMC.
We are in the process of developing a commercial product that will require manufacturing at a rate of 10K to 20K units per month. When we started this work, semiconductor availability wasn't a significant concern. At the start of this year we started to receive quotes with lead times in the order of 52 weeks. Yes, a whole year. Some of the chips we designed into this product became unobtainium.
Well. Not quite. We discovered the supply is plentiful in China. And, of course, pricing is insane. A device that was $0.70 is now over three dollars in China. And, again, of course, the only way you can buy it is if you have your boards manufactured by the supplier. In our case this would mean a contract for 10K to 20K units per month having to leave the prospects of US-based manufacturing --which is already a difficult proposition under normal circumstances.
Never mind roads and bridges. We don't need to touch any of that. We need to dump that money into rebuilding industries and reducing their costs of operation to the bone.
Let me put it this way. Everyone can understand that something like a basic Arduino is a very simple product. In electronics terms, on par with making cloth masks. Well, if you had to manufacture a simple Arduino in the US or Europe today, it would be impossible. Much as is the case for masks, the raw components needed to make such a product are not made in the US and Europe. You might be able to have the raw PCB's made. I would not be surprised if the raw materials for that process were not made in the US and Europe as well (I didn't look into it, so I can't make that assertion). Even if you could make one, the end-user cost would likely be over a hundred dollars per unit (if the components were made in the US and Europe).
That's were we are today. And this isn't going to get any better by fixing roads and bridges.
We need to focus on what is truly important. We needed to do this ten or twenty years ago.
This effort by Intel might not be all it is reported to be. I just hope it is an indication that we are starting to understand what we are standing on...and it isn't solid ground.
I think you have no idea what you're talking about. In Europe at least, there are plenty of fabs that make microcontrollers like the Arduino. And the original Arduino boards (w)are made in Italy which last I checked is still in Europe.
>Much as is the case for masks, the raw components needed to make such a product are not made in the US and Europe.
In Europe we have also plenty of factories making FFP2 masks (my favorite ones I can buy in Austria ATM come from a brand made in the Czech Republic).
For what I remember some(most?) components of arduinos are imported from outside EU. Correct me if I am wrong.
So what? A third of iphone components are imported from outside of china. No competitive economy is 100% self sustainable.
Generally speaking, products with greater complexity illustrate this problem far better. The problem is it would be impossible for the a reader on HN to dig into this reality due to not having access to the bill of materials, etc.
I have never said I am pushing for a 100% localized supply chain. That would be preposterous. Silly. However, the reality we navigate today isn't good at all. Last year showed the world what can happen.
China is nearly at a 100% single-source position. This varies from industry to industry. And, of course, they are not saying "we have enough, let's stop here", they are laser-focused on bringing in as much business as possible and will absorb as much of the supply chain as possible.
What they have done is admirable and nothing less than remarkable. I don't fault or hate them for it. They work hard and deliver results. They deserve to be where they are.
Simple exercise: The bill of materials for Arduino boards is publicly available. Pick a board. Now go source every single component on that board.
Create two spreadsheets. The first would have you source what you can from China. The second would source from any EU country and the US if you'd like and exclude any Chinese made component from the list.
My prediction: You can make 100% of an Arduino in China using Chinese components. You cannot make an Arduino in Europe or the US without Chinese components.
Let's add another layer to this.
Let's say you create a third spreadsheet where you source as many components (and services) as possible from the US and Europe and whatever you can't you get from China.
Now calculate the cost for each case: China only. US/EU only. US/EU + necessary Chinese components and services.
Prediction: The US/EU cost --if it is even possible to do it-- will easily double or triple the cost of that product. The US/EU + China cost won't be competitive at all because the vast majority of the items will have to come from China and you will incur additional costs for producing in the US/EU rather than just facing reality and having the product made entirely in China.
Please note that I am using an Arduino as a super-simple example for the sake of an illustration. It would be wrong to engage in dissecting this simple illustration tool in an attempt to invalidate my claim. This claim comes from decades in manufacturing, where I have faced this question of where I can buy components, assemblies and services. Twenty or thirty years ago sourcing from the US or Europe wasn't horribly difficult. Today, it is frustrating to see just how difficult this is. The problem is real.
Nobody is suggesting 100% repatriation of the supply chain. That is impossible. There is a number between 0% and 100% that would do both the US and Europe much good. I don't know what that number is.
Once you start looking at more "real world" products, the picture gets dark very quickly. Last year we designed a robotic system. Virtually every component that went into making this robot came from China. I think I can say that the aluminum, sheet metal and bearings did not. While I did not conduct a component-by-component survey, I think I can easily say that 95% of the electronic components, connectors, wires, cameras, displays, sensors, pneumatic valves, motors and even the wheels came from China. In other words, if you had to make this robot without China all you'd end-up with is an empty metal shell. That should make everyone take pause.
This is just one layer of this reality. There are many more.
The answer isn't getting rid of China, the answer is to create a competitor so that China cannot screw you over. Instead of letting China become a middle man for goods produced in Africa just go to Africa directly.
There is not much you can do to help private corporations at this point, you've already given them all the support they could possibly get. If nobody bites, that's on them.
Meanwhile low skill workers barely get any support. An infrastructure bill gives them work and keeps the towns and cities from falling apart.
Automotive is sold at n times the markup of commodity chips.
Legacy car manufacturers are basically playing catch up with the likes of Tesla (doing their own chips) or Waymo (nvidia, I believe) to fix their supply chain to not have dozens of poorly integrated black boxes so they can actually update them in a sane way. E.g. VW has been struggling in the last year to get that working with the ID3. They are working with Infineon on chips for this apparently. Other manufacturers are facing similar challenges and are getting there with various levels of success. E.g. if you want to update your Kia EV, you need to go to the dealer.
So, the market opportunity for Intel is making performant SOC aimed at these manufacturers that can keep up with the market leaders and can be used while they figure out to get from level 2/3 autonomy to level 5.
There probably is a market for dumb EVs as well without much intelligence but that also implies low margins. Anything ICE is basically what it is at this point. R&D is basically limited to milking those legacy product lines until they are eventually shut down. So that's not a good market for Intel either.
There are plenty of low margin/low performance manufacturers where Intel stands no chance of ramping up to volumes of sales that would compensate for the low margins. So, high end EV manufacturers with some ambitions on the autonomous driving front is the market for this. Anything else does not make sense for Intel.
The only question is if they keep on insisting on x86. IMHO if they do, they are going to fail, again. Like they failed in the mobile market. The right moment to start targeting this market was 10 to 5 years ago. They are already late.
Really? My major concern with any car I have used ever has never been software.
It still shows the level of desperation there.
1. Intel needs to clear Fab Capacity if they were to move to higher node. This is always a problem in the Fab industry. Normally Intel just scale whatever product line up they have that fits.
2. Car will be the next SemiConductor Battle Ground. It will just be a large computer, lots of sensors with battery moving around. Intel already has MobileEye, they might as well partner with these largest Car manufacture on Chip design for future growth. This is the same play with 5G although Intel seriously messed it up.
3. It also make political wins and help with their change in direction of Custom Foundry or now called Intel Foundry Services.
For me this is a brilliant move. Assuming they execute it perfectly. Some short term pain and lots of long term gain. Which is what I like. A vision and strategy how they are going to win. Something that was lacking with the previous Intel.
Conversely, a lot of the trouble with 10nm and smaller manufacturing is that they shrank the transistors but didn't shrink the dies -- or rather, the ones that did succeed didn't "make a giant 5nm CPU" but they decided to make chiplets instead.
If you really could make the monolithic 5nm part it would crush the chiplets in terms of economics since advanced packaging is more like circuit board assembly than lithography. Maybe they can do it now with new pellicles, but today small geometry parts go good with small dies.
That being said this is a classic "grab headlines while one may, for by the time the deadline rolls around everyone will have forgotten and the goalposts will have moved"
Intel hasn't seriously been in the low margin embedded chip business in decades (margin starts terrible and goes down from there as product matures and you have to guarantee supply for a long time).
Expect a stock bump though.
Intel made 24 billion dollars last year on 78 billion dollars of revenue. It's not exactly near death. 1996 Apple was near death. Intel is having a mid-life crisis at best.
That’s how companies now are evaluated. Grow or gtfo.
Many companies both pay dividends and do stock buybacks, Intel among them.
Profits today, decline tomorrow. Blackberry revenues grew for several years after the iPhone, but they were doomed.
This is Intel’s challenge. Present profits are misleading. The CEO recognizes it and they have a good shot of turning things around though.
Why do I say sickness?
- Most of their revenue comes from a single product line; every BU reported a revenue decline last year except mobileeye (small) and CCG.
- Their parts are not broadly competitive on performance/W nor raw performance except in diminishing subsectors.
- They have no presence in the largest growing sector: mobile.
- Oh, that "eye-popping" $20B investment in fab over the next three years? TSMC plans to invest $100B in the same period!
Now the Intel of old could probably have pulled out of these problems, but let's look how they got here:
- They have historically been a fab powerhouse, with process years ahead of the competition. But they have repeatedly missed their own milestones for smaller node size and are now (relatively behind or even far behind). They've had to throw in the towel repeatedly over the past few years in this area.
- They've spent time fiddling at the edges (e.g. depth cameras) pouring money into new businesses and then dropping them. Not a bad idea when you need growth and should diversify your revenue stream (look at my first point above) but terrible if you have a 100% whiff rate.
- Of those whiffs above they have had some bad ones, like radio chipsets, when they earlier had said they were counting on them.
- Their architecture appears to be inappropriate for mobile/low power. This is a self-own. They not only had for its time the fastest ARM product line (StrongARM, from the DEC acquisition) but threw it away because they were focused on their high-margin quasi-monopoly x86 line.
- Their partners (Microsoft, Dell, etc) are enthusiastically exploring ARM and Intel at the moment has no counter.
- Their "only the paranoid survive" culture is gone; business decision making has been slow and lazy for more than a decade.
- They are heavily dependent on their partners. If there is a phase change in the market (MS finally ships a decent ARM Windows, Dell produces a decent ARM laptop that runs it), well, bye bye a big chunk of CCG revenue. They could save some share by slashing price, but it's still terrible.
This is why I said it's near death.
Does Intel have a problem? Yes. They know it too. I think you're overlooking how long a runway $20 billion buys you though to correct things & try different strategies. It's also ignoring the huge push within the US government to subsidize silicon manufacturing in the US for political & security reasons, of which Intel will be a huge beneficiary. I fully expect in 15 years Intel will still be a very valuable independent company with a market cap larger than today.
[1] https://www.macrotrends.net/stocks/charts/INTC/intel/cash-on... [2] https://money.cnn.com/2002/04/12/pf/agenda_msft/
Intel's sales have increased every one of the past 4 years (which is all the data this site shows): https://www.marketwatch.com/investing/stock/intc/financials/...
I dont think Intel is necessarily going to have an easy next few years, but they are still a massive company who is growing every year.
Your rose coloured glasses come from survivorship bias. Apple was a magical story, true. AMD made it but is a tiddler by comparison. Based on history, they’re going to need abnormally good luck.
More to the point: I would like Intel to succeed but they face long odds. Looking at localized data like top line revenue ignores secular shifts that have been moving against them, and to which they have so far been demonstrably unable to respond.
Looking at just the top line is like deciding how your journey is doing based on your current speed. It’s an input but tells you almost nothing on its own.
I demand satisfaction, sir! Emails at twenty paces. No IRC allowed.
Maybe it could make a great blockchain/nft startup. DuelUp, put your money where your mouth or in this case (words in text) are.
heh
Not as clumsy or random as HTML or MIME; an elegant weapon for a more civilized age.
I don't see this as them getting into low margin chips. This is them selling manufacturing capacity to someone who wants what in ordinary times might be low margin chips. And some of them might not actually be all that low margin; I assume modern cars have one or two computers that are more like smart phones than Arduinos.
I worked with Toyota when they were literally trying to shave a ¥100 from the engine control unit of a new vehicle. A new design! Once they entered production they were going to shave the cost even more on the backs of their suppliers.
There's a reason why Intel got out of this business. Ever wonder why AMD can use x86? They got that license back when nobody would design in your part unless there was a second supplier of the identical part. So they got a license so that Intel could sell parts. Intel abandoned that world without a backwards glance.
Nothing like the possibilities of massive economies of scale. There's a reason Toyotas are such good values.
Toyotas are not "good value" because they're cheap -- just look at the prices for new Toyota Tacomas! They're a good value because of the exceptional engineering, and they sell at a premium because of it.
People need to stop drinking this stupid "scale is all that matters" kool-aid.
AMD can use x86_64 because they invented it.
Intel most definitely invented x86 since the x86 is itself a reference to the Intel 8086.
This is such a perfect description, it make me want to cry.
On Intel's side, they desperately need to prove that they can have an amicable relationship with a customer if they have any hope of making their IDM 2.0 model work. This will help them build out the relationships they need, assuming they don't screw it up.
Sounds like a win-win, but I'm guessing the next thing we will hear is a pitch for federal subsidy for this under a national security banner.
That's the narrative (re: dependence on foreign chips), but it's not what actually happened. Last year the auto manufacturers anticipated a slowdown in demand and cut back on inventory orders. This made sense at the time (to reduce orders) given the uncertainty but it sounds like they pretty much cancelled them. They likely figured that their suppliers would be happy for the business when they returned.[1] The semi companies just put other customers in line ahead of them and now the autos are screwed since they are facing incredibly long lead times for chips and don't have the inventory to bridge the gap to make the number of vehicles they need.
This benefits Intel because the line of customers looking for 14nm chips at the prices Intel probably wants to charge likely isn't very long. However, they've found a perfect (i.e. desperate) customer in the autos. So it's actually a win-win for the time being.
[1] This actually works if you're a supplier exclusively to one company/industry. It doesn't work so well if you have other options.
Unless there's a fire at an AKM plant. Or a fire at a Renesas plant. Or a drought in Taiwan. Or blackouts in Texas.
When you're running at low utilization, lead times are shorter and the market is more forgiving of order alterations and incorrect predictions of demand. But that's not the case right now, and there have been several exogenous setbacks for the sector.
Regarding utilization rates, that's probably where most of my snarkiness was coming from: the autos knew demand was high before they cancelled their orders and they chose to do so anyways. I'm sure they had every right to do what they did (contractual language allowing for cancellations etc), but when you step out of line for whatever reason it should come as no surprise that when you decide that you want to be in line after all that you're going to now be at the back of it. This has nothing to do with an unpredictable supply chain and everything to do with an unpredictable customer who didn't plan properly. Yet the story gets played up as the failure of a brittle supply chain.
Just open your ears and you'll hear it already: "Intel CEO calls for "moonshot" to boost U.S. role in chipmaking" https://www.axios.com/intel-ceo-gelsinger-chipmaking-moonsho...
Nom nom!
e.g., Under new management.
But I suspect that ship is sailing regardless.
If not philips somebody like philips.
https://www.asml.com/en/company/about-asml/history
29% ROI Year to date. nearly 500% on 5 years.
Intel I'm assuming did a brainstorm relatively recently and wrote down on a white board "what should we be focusing on".. and they probably wrote down:
-cloud server infrastructure
-something to compete with the Apple M series processors
-high margin businesses ( cars / self-driving and military / large super computer installs )
So basically my understanding is that they use the lead customer in each product segment they care about to help them define the cutting edge for that space. So they probably will use Microsoft/Azure for cloud customer leadership, some competitor of Tesla like VW or Mercedes for ADAS stuff, and military they probably are already doing fine there. Especially with the new Arizona fabs.
Whether or not that's plausible or a marketing play, I'm not sure (since nobody outside of Intel really knows the details about Intel's process constraints), but it doesn't sound like they are targeting an area of competition where they would be introducing new silicon and expecting a whole iteration lifecycle. The intent seems to be drop-in replacements.
with that customer naturally describing a faster horse that it wants :)
A gorilla like Intel will be able to saturate the market few times over if we take all of their capacity.
You can see where consumers could be confused since they basically made a word mean the opposite of what it means. Someone needs a paddling for that.
- Very, very attractive margins on the auto chips.
- Much lower expected margins on Intel chips or idle capacity.
So which is more likely?
(and of course Intel quickly establishes a position in a new market in line with PG's strategy and gets some good PR).
Never mind that by the time Intel actually delivered the other players would have been at the same tech level.
or
Intel’s commitment to a “mixed foundry-proprietary” business model.
the 2013 BMW ... if you got the big headunit you had to also get the big AC... why?
there was an extra hose from the AC to the headunit... why?
intel ATOM. that's why.
Oh, you want auto-park? We bundle that with our premium loudspeakers, it's the "advanced technology pack". Keyless entry? We bundle that with the heated steering wheel.