Intel and Apple tipped to be the first to adopt TSMC's N3 node-based chips

jsilva

Posts: 176   +1
Staff
In brief: Although we are still about a year away from seeing TSMC 3nm process node being mass-produced, reports about which companies will adopt it have already started to emerge. As it seems, Intel and Apple will be the first to leverage the N3 node, with Apple being the first to release a device powered by it, the next-gen iPad.

Apple has been suggested as one of the first companies to take advantage of the N3 node after reports that it was partnering up with TSMC in the node's risk production, expected to begin later this year. Once the N3 node's mass production begins, scheduled for the second half of 2022, Intel will reportedly join Apple as one of TSMC's first customers to benefit from it.

In Apple's case, it's said that TSMC's N3 node will be present on the next generations of iPad devices, which are also expected to be the first devices powered by this node. As for Intel, the company confirmed TSMC will be present in its 2023 product lineup without specifying which technology will be employed. However, rumors claim it will be featured on laptop and server CPU architectures.

"Currently the chip volume planned for Intel is more than that for Apple's iPad using the 3-nanometer process," stated one of Nikkei Asia's sources. Commercialization of N3 node-based chips is expected to begin in the second half of 2022, so products using them should come out in late 2022 or early 2023.

Compared to TSMC's 5nm process node, currently used to power Apple's M1 chip, the N3 node will offer 10% to 15% more computing performance or reduce power consumption by up to 30%. TSMC is also developing an N4 node, which some sources claim will be used on the next generation of iPhone devices.

Also Read: Apple M1: Why It Matters

Following Intel and Apple, AMD and Huawei should also join the list of customers to use TSMC's 3nm process node-based chips, but only later down the line, once the process is more "mature."

Image credit: Laura Ockel

Permalink to story.

 

Irata

Posts: 1,675   +2,812
AMD doesn't want to pay high price for non-mature process (=low yields) when price is per wafer. Intel and Apple can afford it.
Normally there isn‘t enough capacity for AMD on TSMC‘s process as Apple eats it all up.

But if you were TSMC, would you strengthen one of your best customer‘s competitor for a short term gig, risking long term business? With AMD, they can work on new manufacturing tech - doing the same with Intel, a fab competitor, would be stupid.
 

HardReset

Posts: 1,240   +898
Normally there isn‘t enough capacity for AMD on TSMC‘s process as Apple eats it all up.

But if you were TSMC, would you strengthen one of your best customer‘s competitor for a short term gig, risking long term business? With AMD, they can work on new manufacturing tech - doing the same with Intel, a fab competitor, would be stupid.
Apple eats up all capacity on new nodes as they can afford low yields. When node becomes mature, Apple gets another new node and abandon mature node mostly.

This solves one problem for TSMC: Apple would pay for new node, but Apple uses so small chips that amount of produced wafers is quite small. It also means pretty slow ramp up for new node. Now, when Intel pays for "test drive" and wants much more wafers than Apple, TSMC could ramp up 3nm much faster. That benefits TSMC and AMD. So yes, I would. What Intel benefits from this is much better question...
 

captaincranky

Posts: 17,211   +5,964
All I see is Intel and Apple paying for, and getting sucked in by, a massive over abundance of TSMC's ambitions.

The fact of the matter is, TSMC can't deliver enough wafers to meet demand at 7 nm. So, all of a sudden we're at 3 nm?

Automobile plants are shut down for lack of chips, and everybody seems to be whimpering about video cards selling for double list price.

Neither TSMC nor Samsung will have their US fabs finished until likely 2023. And those will likely have to catch up with what's happening now with chip shortages.

So, call me crazy but, TSMC appears to be full of sh!t at 7nm. One has to wonder how much more waste they'll retain, when they claim, "3 nm is just around the corner"?
 

Irata

Posts: 1,675   +2,812
Apple eats up all capacity on new nodes as they can afford low yields. When node becomes mature, Apple gets another new node and abandon mature node mostly.

This solves one problem for TSMC: Apple would pay for new node, but Apple uses so small chips that amount of produced wafers is quite small. It also means pretty slow ramp up for new node. Now, when Intel pays for "test drive" and wants much more wafers than Apple, TSMC could ramp up 3nm much faster. That benefits TSMC and AMD. So yes, I would. What Intel benefits from this is much better question...
Intel would be able to offer chips on a competitive node and no, that does not benefit AMD.

One of Intel‘s competitive advantages historically was being ahead of their competition wrt processes. Not sure how getting back that advantage would benefit AMD or TSMC - Intel is also a fab competitor.

And if there is excess capacity on 3nm, I doubt that TSMC would not find anyone else besides Intel to snap that up. This simply does not make sense from a business perspective unless there is an angle to this deal that has not been disclosed yet.
 

HardReset

Posts: 1,240   +898
All I see is Intel and Apple paying for, and getting sucked in by, a massive over abundance of TSMC's ambitions.

The fact of the matter is, TSMC can't deliver enough wafers to meet demand at 7 nm. So, all of a sudden we're at 3 nm?
TSMC builds new fab for 3nm, so 3nm capacity does not reduce 7nm capacity at all...
Automobile plants are shut down for lack of chips, and everybody seems to be whimpering about video cards selling for double list price.

Neither TSMC nor Samsung will have their US fabs finished until likely 2023. And those will likely have to catch up with what's happening now with chip shortages.

So, call me crazy but, TSMC appears to be full of sh!t at 7nm. One has to wonder how much more waste they'll retain, when they claim, "3 nm is just around the corner"?
Automobile manufacturers decided not to order chips because of pandemic. Then they were quite surprised that chip manufacturers sold their capacity to someone else. No problems there.

Because miners buy all cards they can get, AMD and Nvidia do not want to sell too much cards as they know what happen when cryptomining is no longer profitable. AMD could manufacture more cards as could Nvidia, but for obvious reason they don't want to.

There is no chip shortage at all. It's just chips are going for wrong place (scalpers, miners...) or buyers were just stupid (automobile manufacturers). For video card prices, Nvidia is one to blame. Too low price for 3000-series. Scalpers noticed that price cannot go any lower so they can buy everything and sell for profit without any fear of making loss. Same applies to consoles.

There are no problems with TSMC 7nm. Like I said before, GlobalFoundries exit caused Some 7nm capacity shortage. But currently there is not any.
Intel would be able to offer chips on a competitive node and no, that does not benefit AMD.

One of Intel‘s competitive advantages historically was being ahead of their competition wrt processes. Not sure how getting back that advantage would benefit AMD or TSMC - Intel is also a fab competitor.
That does not benefit AMD but faster ramp up will help much more than Intel getting some chips on semi competitive node (bad yields, bad clocks etc).

Again, TSMC (and AMD) wants someone to pay for new process test drive. Apple wants so few wafers that ramp up takes ages. Just look at 5nm: risk production started March 2019 (2 years, 4 months ago), mass production started April 2020 (1 year, 3 months ago), real ramp up coming Q4/2021, that is, over 2.5 years after risk production and 1.5 years after mass production. AMD 5nm products are coming probably Q2/2022. That is, over two years after mass production started. I consider that ultra slow. One reason is low capacity when mass production started.
And if there is excess capacity on 3nm, I doubt that TSMC would not find anyone else besides Intel to snap that up. This simply does not make sense from a business perspective unless there is an angle to this deal that has not been disclosed yet.
Anyone else, like who? Who wants to pay huge amount of money for crappy process (low yields, bad clocks) AND order thousands of wafers. Looking past, TSMC had only found Apple. It does make perfect sense to grab another test driver.
 

Burty117

Posts: 4,193   +2,212
Yes, but a smart business gives preferences to long term customers (Apple, AMD) vs. short term gigs even if they pay well.
Normally there isn‘t enough capacity for AMD on TSMC‘s process as Apple eats it all up.

But if you were TSMC, would you strengthen one of your best customer‘s competitor for a short term gig, risking long term business? With AMD, they can work on new manufacturing tech - doing the same with Intel, a fab competitor, would be stupid.
Ok, where would Intel / AMD / Apple go? Seriously, TSMC are like, the only ones who have managed to get to 7nm and below, all other fabs haven't managed to surpass 8nm (that I'm aware of).
 

NeoMorpheus

Posts: 883   +1,678
Ok, where would Intel / AMD / Apple go? Seriously, TSMC are like, the only ones who have managed to get to 7nm and below, all other fabs haven't managed to surpass 8nm (that I'm aware of).
One item I keep bumping into, apparently, tsmc process is not truly 7 or 5 or whatever nm.

there seems to be a density factor that they are “misleading “ the rest with.

old discussions here:

https://www.reddit.com/r/hardware/comments/bx9408
 

Burty117

Posts: 4,193   +2,212
One item I keep bumping into, apparently, tsmc process is not truly 7 or 5 or whatever nm.

there seems to be a density factor that they are “misleading “ the rest with.

old discussions here:

https://www.reddit.com/r/hardware/comments/bx9408
Oh I know all about this, the "Xnm" is more of a marketing turn than actual fact. My point was though, where else is Intel / Apple / AMD going to go for "3nm". I guess Samsung being the only real competitor?
 

Irata

Posts: 1,675   +2,812
Oh I know all about this, the "Xnm" is more of a marketing turn than actual fact. My point was though, where else is Intel / Apple / AMD going to go for "3nm". I guess Samsung being the only real competitor?
They are pushing hard to get some of that sweet government money, so TSMC may just be a temporary gig.

 

HardReset

Posts: 1,240   +898
IIRC, AMD has the entire 5nm node locked-up which is why Apple and Intel are looking down the road at 3nm.
AMD has so far released following 5nm products:

Apple has released quite many. Apple pays for getting 5nm first and pays for getting 3nm first. Simple as that.

AMD has locked Some capacity for TSMC 5nm node. Not everything, that's 100% certain. AMD just cannot afford it.
 

Avro Arrow

Posts: 1,727   +2,046
TechSpot Elite
AMD has so far released following 5nm products:

Apple has released quite many. Apple pays for getting 5nm first and pays for getting 3nm first. Simple as that.

AMD has locked Some capacity for TSMC 5nm node. Not everything, that's 100% certain. AMD just cannot afford it.
You could be right, I just remember that AMD had a significant investment in it.
 

HardReset

Posts: 1,240   +898
You could be right, I just remember that AMD had a significant investment in it.
Of course they have significant investment. But AMD buying all TSMC 5nm production for two years would cost at least around 40-50 billion. AMD's yearly revenue is around 10-15 billion.
 

Irata

Posts: 1,675   +2,812
Of course they have significant investment. But AMD buying all TSMC 5nm production for two years would cost at least around 40-50 billion. AMD's yearly revenue is around 10-15 billion.
TSMC’s total revenue for 2020 was $47 Billion, so I doubt that number for a single node over two years.

However, given AMD’s current product stack, they can probably sell every last die they make, so having to spend more would not be an issue since they‘d also earn more.
 

HardReset

Posts: 1,240   +898
TSMC’s total revenue for 2020 was $47 Billion, so I doubt that number for a single node over two years.
According to reports (like https://www.digitimes.com/news/a20210422VL200.html ) TSMC will produce 150K wafers per month on 5nm tech. Pricing for single wafer is unknown, rumours say around 18K$. Considering how difficult process is and how transistor density has improved, that might be fairly correct. To be safe, let's assume wafer price is "only" 12K$ (around same as 7nm):

12K$*24 months*150K wafers per month = 43.2 billion $. Right, if that 18K$ holds true, TSMC will almost match it's revenue vs 2020 just with 5nm production. Assuming 150K wafers per month is true also.
However, given AMD’s current product stack, they can probably sell every last die they make, so having to spend more would not be an issue since they‘d also earn more.
AMD's product stack is quite narrow right now and I very much doubt AMD could sell that much. Still, AMD has shortage of some low demand chips, some big demand but big die area chips and some products (mid range 7nm GPU's) are not even announced because of manufacturing capacity.
 

Burty117

Posts: 4,193   +2,212
What about Samsung???
It's a good point, Samsung do have a 5nm fab, so your choices are TSMC or Samsung.

I don't know enough about either firms chip making facility's to judge who's is better or worse, what I can say is though, I haven't been able to purchase a 3000 series GPU at all, at all.
I have been able to get hold of AMD's 5000 series CPU's and I have managed to get hold of a PS5 (with a lot of work and patience mind you, it wasn't easy getting hold of either) of which, both are manufactured by TSMC.

So if I was going based on how much Samsung can meet demand vs TSMC, I'd wager TSMC is the safer bet.

And from what I've been reading, Nvidia aren't really competing over at Samsung for fab space either as they're using some customised 8+nm fab specifically for themselves.

Of course, it could simply be demand really was that extraordinarily high for Nvidia GPU's.
 

captaincranky

Posts: 17,211   +5,964
So if I was going based on how much Samsung can meet demand vs TSMC, I'd wager TSMC is the safer bet.
Maybe, maybe not. As I mentioned earlier, Samsung is building a 17 billion dollar fab in Texas, and TSMC is springing for a 12 billion plant, in Arizona.

IMHO, when those two fabs are completed, it's anybody's guess as to who'll come out on top.

As it stands now, it would appear that TSMC's promises are better than Elon Musk's, but not by a whole helluva lot.