Reading between the lines - MediaTek will probably send Intel some obscure parts for a small production run. This is "Intel 16" we're talking about, which is revised 22nm FinFET Low Power (that Intel thinks is comparable to 3rd-party 16nm). Aka Smart Home device chips, nothing near smartphone level.
So... they might have extra capacity, sure, but extra capacity isn't the problem right now. NVIDIA has overbooked and is desperately seeking a buyer for capacity when everyone is cutting back or has all the capacity they need. This won't affect the 5nm-ish capacity, but only TSMC 22nm capacity, which... how many buyers does that have? Is it overbooked?
Isn't this right around the node that is under huge pressure right now, causing unprecedented shortages and backlogs for sectors such as automotive or medical devices? Might be nice to have more help there.
This was a major point of contention during the chip shortages. Automotive and medical devices are stuck on very old nodes, like 180nm. This MediaTek partnership is using way, way too new of a node than they would like, while being too old for smartphones. 22nm would be lightyears ahead of where they are.
Chip makers view the situation as being simple: Car makers and Medical need to upgrade their tech to newer nodes, as they aren't interested in building new 180nm-era factories. Car makers and Medical chip makers blame, well, the chip makers and say that redesigning those chips would be very expensive as well as to re-certify, but so far the chip makers don't care.
And so rumor has it, though rumor only for now, that car makers are finally updating because there is no other choice. To what though? Probably the next-oldest and cheapest node, 40nm or older, not 22nm.
I’ve heard many people repeat that summary, but as you appear to have first hand knowledge: what do you believe or know to be the reason for the shortage?
The shortage was the car industry cancelled all there orders for chips (mistakenly, because of bad forecasting) and the chip manufacturers sold that factory time elsewhere, cars tried to take it back and buy more factory time to make up for the shortfall… but they all did it. So it was basically like toilet paper shortages.
You are right they are on old architecture and the chip manufactures would prefer to move to newer ones. But I don’t think they ever didn’t take the car industry seriously, the order sizes are a stupid amount of money to leave on the table over “new nodes vs old nodes”
The problem with "new nodes vs. old nodes" for the chip manufacturers is that any semiconductor plant has quite large fixed costs.
When the production is very high, the fixed costs are divided over a very large number of chips and the cost per chip is negligible.
When the production diminishes for an old node because some obsolete devices are no longer used in new designs, there is a threshold where the fixed costs per chip become so large that the chips cannot be sold at a profit.
So even if the semiconductor plant would still have orders for millions of chips per month and they could be easily made, the production would not be profitable, because the profitability threshold might be e.g. at ten million chips per month.
Because of that, as soon as they see that the old plants are run much below their maximum capacity, the foundries would prefer very much the customers to move their orders to a newer node, to be able to close completely the old plant instead of running it at a very low profit or even at a loss, when they have to satisfy some old contracts, made when the plant was still profitable.
That may be true. I can believe that. But the car industry is monstrously big, each car having 10s-100s of chips. There isn’t really a comparable volume of product that would make it not worth servicing them.
I agree you might try and use this opportunity to shift the OEMs onto chips that match other more modern clients… but it’s tail wagging the dog territory.
Every single manufacturer in the car industry collectively cancelled their orders?
Tesla, for instance, would’ve had completely different forecasting as an EV-only manufacturer with multiple new models and lines coming to the market. I have doubts that they would have scaled back orders.
They didn’t, but they got caught up in being unable to scale up as their production scaled up. So they were still buying toilet paper at the normal rate, but suddenly all the toilet paper disappeared
Edit: agree it is super weird the entire industry (or at least enough of a critical mass) made the same decision on their Covid forecasting… it’s almost unbelievable/conspiracy territory. But it is what happened.
Isn't the car industry consolidated to 4 or 5 big groups? It's just a Zoom call or a small table in a Davos chalet to have the entire industry making the same decision, no conspiracy required.
Q4 ish. It’s returning to normal now in the factories, but there is a back log of orders and logistics that will delay this to consumers for another 6 months minimum
I mostly agree with what you have said, except that I believe that anyone who updates their older digital devices to a newer node now would be wise to choose 28 nm instead of 40 nm or of any other older node, unless constrained by suitability of the node for certain analog parts of the devices.
The 28 nm node is the last before FinFET, so its costs do not differ much from 40 nm, but such a choice would be much more future proof.
Already now the production capacity for 28 nm is greater than for any older nodes. Because more 28-nm plants are planned to be built, the difference in capacity will become even greater in the next years.
Choosing 40 nm now might force another process upgrade after not many years.
EDIT:
After writing above based only on the existing and planned production capacities, I have googled and I have found that representatives of TSMC have made public statements that no new plants for nodes older than 28 nm will be built, so everybody who upgrades must choose at least 28 nm (which was state-of-the-art 10 years ago):
While this statement is from TSMC, it is quite certain that also none of the other foundries will build new plants for nodes older than 28 nm, because they would not be able to recover their investments.
Would it not make sense to upgrade to the newest node they can afford to a longer lifespan more life out of their next node? It feels like picking the next-oldest node leaves them having to solve this issue again in the near future...
Mediatek accounts for 6% of TSMC revenue, meaning that they aren't such a large a customer either in advanced nodes, volume, or both. Some capacity may be freed, but I doubt that it would be significant enough for smartphone chips, especially given that Mediatek is going for an older node size on Intel.
Given Intel’s execution track record and even recent mishaps (where’s my Sapphire Rapids Pat, or the gpus?!) even getting this is a big step. They better deliver their A game if they hope to win any real trust.
20 comments
[ 0.22 ms ] story [ 55.6 ms ] threadSo... they might have extra capacity, sure, but extra capacity isn't the problem right now. NVIDIA has overbooked and is desperately seeking a buyer for capacity when everyone is cutting back or has all the capacity they need. This won't affect the 5nm-ish capacity, but only TSMC 22nm capacity, which... how many buyers does that have? Is it overbooked?
Chip makers view the situation as being simple: Car makers and Medical need to upgrade their tech to newer nodes, as they aren't interested in building new 180nm-era factories. Car makers and Medical chip makers blame, well, the chip makers and say that redesigning those chips would be very expensive as well as to re-certify, but so far the chip makers don't care.
And so rumor has it, though rumor only for now, that car makers are finally updating because there is no other choice. To what though? Probably the next-oldest and cheapest node, 40nm or older, not 22nm.
Do you have a source for this?
I work with most of the OEMs and “update to newer chips” is not something they are doing.
Tesla was effected by the shortage as well even though they expressed willingness to use whatever chip design they could get their hands on.
I don’t think your summary is accurate. I could be wrong but my first hand experience tells me the exact opposite.
You are right they are on old architecture and the chip manufactures would prefer to move to newer ones. But I don’t think they ever didn’t take the car industry seriously, the order sizes are a stupid amount of money to leave on the table over “new nodes vs old nodes”
When the production is very high, the fixed costs are divided over a very large number of chips and the cost per chip is negligible.
When the production diminishes for an old node because some obsolete devices are no longer used in new designs, there is a threshold where the fixed costs per chip become so large that the chips cannot be sold at a profit.
So even if the semiconductor plant would still have orders for millions of chips per month and they could be easily made, the production would not be profitable, because the profitability threshold might be e.g. at ten million chips per month.
Because of that, as soon as they see that the old plants are run much below their maximum capacity, the foundries would prefer very much the customers to move their orders to a newer node, to be able to close completely the old plant instead of running it at a very low profit or even at a loss, when they have to satisfy some old contracts, made when the plant was still profitable.
I agree you might try and use this opportunity to shift the OEMs onto chips that match other more modern clients… but it’s tail wagging the dog territory.
Tesla, for instance, would’ve had completely different forecasting as an EV-only manufacturer with multiple new models and lines coming to the market. I have doubts that they would have scaled back orders.
Edit: agree it is super weird the entire industry (or at least enough of a critical mass) made the same decision on their Covid forecasting… it’s almost unbelievable/conspiracy territory. But it is what happened.
The 28 nm node is the last before FinFET, so its costs do not differ much from 40 nm, but such a choice would be much more future proof.
Already now the production capacity for 28 nm is greater than for any older nodes. Because more 28-nm plants are planned to be built, the difference in capacity will become even greater in the next years.
Choosing 40 nm now might force another process upgrade after not many years.
EDIT: After writing above based only on the existing and planned production capacities, I have googled and I have found that representatives of TSMC have made public statements that no new plants for nodes older than 28 nm will be built, so everybody who upgrades must choose at least 28 nm (which was state-of-the-art 10 years ago):
https://www.anandtech.com/show/17470/tsmc-to-customers-time-...
While this statement is from TSMC, it is quite certain that also none of the other foundries will build new plants for nodes older than 28 nm, because they would not be able to recover their investments.