3nm Chips From Apple Supplier TSMC to Enter Mass Production This Week

[BorisDG]

It's well-known, from TSMC's own press releases, that the N4P process (likely used in the A16) is 5 nm, not 4 nm:

TSMC Expands Advanced Technology Leadership with N4P Process|Taiwan Semiconductor Manufacturing Company Limited

Hsinchu, Taiwan, R.O.C., Oct. 26, 2021 - TSMC (TWSE: 2330, NYSE: TSM) today introduced its N4P process, a performance-focused enhancement of the 5-nanometer technology platform.

pr.tsmc.com

"As the third major enhancement of TSMC’s 5nm family, N4P will deliver an 11% performance boost over the original N5 technology and a 6% boost over N4."

So how much weight should we give to DigiTimes' TSMC reporting when they aren't even aware of something so basic?

And why doesn't MR do some basic fact-checking before posting this stuff? Given that the whole point of this article is to report on a process change, you'd think they want to get the process descriptions right.


This doesn't make sense. First they say that mass production on the first gen 3 nm chip (N3) starts in a few days, but now they're saying that production on 3 nm is unlikely to ramp up until the next gen 3 nm (N3E) gets underway. This is just screaming out for clarification, yet MR gives us none. Come on guys, how putting some thought into these articles, rather than just cutting and pasting a few quotes from another source (which, incidentally is behind a paywall, preventing us from checking it ourselves)?

It is 4nm just like Snapdragon 8 Gen 1 Plus was or Mediatek Dimmensity. 5 -> 4 -> 3

Its all over 4nm.

 

[theorist9]

If only the facts supported your argument.

To quote from TSMC's own website: https://www.tsmc.com/english/dedicatedFoundry/technology/logic/l_5nm#l_5nm

"In addition, TSMC plans to launch 4nm (N4) technology, an enhanced version of N5 technology. N4 provides further enhancement in performance, power and density for the next wave of N5 products."

Yes, N4 is widely considered part of the N5 family, including by TSMC. No, you are wrong to say it's "5nm", as that's just a label, and the one chosen by TSMC is "4nm".

In fact I wouldn't be at all surprised to see TSMC refer to it as 5nm in other places on their web site. Either way, my point is proven: To declare it to *not* be 4nm is silly. It's as much 4nm as N5 is 5nm. (Which is to say, not at all, really, but pay no attention to that man behind the curtain. He's only 4.5nm tall, after all.)

Yes, I recall that. TSMC corrected that misinformation in the link I quoted in my original post on this:

"As the third major enhancement of TSMC’s 5nm family, N4P"

TSMC Expands Advanced Technology Leadership with N4P Process|Taiwan Semiconductor Manufacturing Company Limited

Hsinchu, Taiwan, R.O.C., Oct. 26, 2021 - TSMC (TWSE: 2330, NYSE: TSM) today introduced its N4P process, a performance-focused enhancement of the 5-nanometer technology platform.

pr.tsmc.com

The fact is that saying N4P is 4 nm misrepresents it as being on a different process (which it is not), rather than an optimization of an existing process (which it is). Do you understand?

See here for more info:

TSMC’s 3nm Conundrum, Does It Even Make Sense? – N3 & N3E Process Technology & Cost Detailed

Shrinking finally costs more, Moore’s Law is now dead in economic termsA couple of weeks ago, we were able to attend IEDM, where TSMC presented many details about their N3B and N3E, 3nm class…

www.semianalysis.com

 

[Confused-User]

Yes, I recall that. TSMC corrected that misinformation in the link I quoted in my original post on this:

"As the third major enhancement of TSMC’s 5nm family, N4P"

TSMC Expands Advanced Technology Leadership with N4P Process|Taiwan Semiconductor Manufacturing Company Limited

Hsinchu, Taiwan, R.O.C., Oct. 26, 2021 - TSMC (TWSE: 2330, NYSE: TSM) today introduced its N4P process, a performance-focused enhancement of the 5-nanometer technology platform.

pr.tsmc.com

The fact is that saying N4P is 4 nm misrepresents it as being on a different process (which it is not), rather than an optimization of an existing process (which it is). Do you understand?

TSMC did *not* correct that "misinformation" - it's still on their web site.

Further, there is no reason why "4nm" (which means nothing) shouldn't be a part of the "process family" started with a "5mn" (also meaning nothing) node. So your quote is not a correction.

"Xnm" is semantically null, except possibly to indicate that "Xnm" is better than "Ynm" if X < Y, and FSVO "better".

Your insistence on placing importance on that term is misleading you.

 

[saberone]

It's well-known, from TSMC's own press releases, that the N4P process (likely used in the A16) is 5 nm, not 4 nm:

TSMC Expands Advanced Technology Leadership with N4P Process|Taiwan Semiconductor Manufacturing Company Limited

Hsinchu, Taiwan, R.O.C., Oct. 26, 2021 - TSMC (TWSE: 2330, NYSE: TSM) today introduced its N4P process, a performance-focused enhancement of the 5-nanometer technology platform.

pr.tsmc.com

"As the third major enhancement of TSMC’s 5nm family, N4P will deliver an 11% performance boost over the original N5 technology and a 6% boost over N4."

So how much weight should we give to DigiTimes' TSMC reporting when they aren't even aware of something so basic?

And why doesn't MR do some basic fact-checking before posting this stuff? Given that the whole point of this article is to report on a process change, you'd think they want to get the process descriptions right.


This doesn't make sense. First they say that mass production on the first gen 3 nm chip (N3) starts in a few days, but now they're saying that production on 3 nm is unlikely to ramp up until the next gen 3 nm (N3E) gets underway. This is just screaming out for clarification, yet MR gives us none. Come on guys, how putting some thought into these articles, rather than just cutting and pasting a few quotes from another source (which, incidentally is behind a paywall, preventing us from checking it ourselves)?

Yes. Weird “news”. Earlier this year it was reported that TSMC would abandon N3 for N3E.

Plz clear it up.

 

[Confused-User]

Yes. Weird “news”. Earlier this year it was reported that TSMC would abandon N3 for N3E.

Plz clear it up.

You overstate. The original process ("N3B") did not meet some targets, and most customers are expected to start with N3E, which has a number of benefits over N3B but at the price of somewhat relaxed design rules (it's not as dense). But N3B wasn't abandoned.

Apple is still widely believed to be the tentpole customer for N3B, though I expect they'll use N3E for the A17. If that's true, then it's a foregone conclusion that they're using N3B for *something* in between now and then, which would almost certainly be the larger MacBook Pros, iMac(s), Studio, and possibly Mini. Which puts shipping for those products sometime between late March and mid-April, probably, with announcements possibly a few weeks earlier.

 

[theorist9]

TSMC did *not* correct that "misinformation" - it's still on their web site.

Further, there is no reason why "4nm" (which means nothing) shouldn't be a part of the "process family" started with a "5mn" (also meaning nothing) node. So your quote is not a correction.

"Xnm" is semantically null, except possibly to indicate that "Xnm" is better than "Ynm" if X < Y, and FSVO "better".

Your insistence on placing importance on that term is misleading you.

They did issue a correct statement, just didn't clean up what the clueless marketers wrote on the page you linked.

It's long been standard practice in the industry to reserve a new nm designation for an entirely new process technology. That's what you don't understand. Dylan Patel makes that clear in the article from semianalysis I linked earlier, explaining that N4P is not a new process technology, it's just an optimization of the 5 nm node. He calls it a "nodelet". You should read that article. It would help you to understand things better.

 

[Confused-User]

They did issue a correct statement, just didn't clean up what the clueless marketers wrote on the page you linked.

It's long been standard practice in the industry to reserve a new nm designation for an entirely new process technology. That's what you don't understand. Dylan Patel makes that clear in the article from semianalysis I linked earlier, explaining that N4P is not a new process technology, it's just an optimization of the 5 nm node. He calls it a "nodelet". You should read that article. It would help you to understand things better.

Lol. I check Semianalysis fairly regularly, and read that article a few days ago. I didn't need to read it to know that N4 is a die shrink of N5. So what?

You persist in treating the "nm" designations as having significant meaning. Until you realize that they don't, you will waste your time on meaningless trivialities. I however am done... feel free to have the last word if you like.

 

[Allen_Wentz]

Well it sucks to be an iPhone 14 Pro owner. USB-C and 3nm for the iPhone 15 Pro. 👌

Nonsense. The iPhone 14 Pro rocks. We know that there will always be better tech coming next year, duh.

 

[theorist9]

Oh Jesus Christ. Do you also want to tell us the breaking news that nothing on N5 is actually sized at 5nm, and nothing on N3 is sized at 3nm?
Look that ship has sailed. Anyone who hasn't been dead for the past ten years is WELL AWARE that N5 means '5nm class" which means fsckall 5nm related except that it's a set of improvements that give you about 1.8x density increase over N7.
Likewise anyone who actually matters and cares is well aware that N4 is an optical shrink of N5. You don't need to act like you're Woodward and Bernstein here revealing some scandal.

You really need to work on your reading comprehension. Accusing me of acting like Woodward and Bernstein is a straw man, because that is precisely the opposite of what I was doing. I clearly said the reason DigiTimes and MR should have gotten this right is because it is well-known (at least well-known to the tech press), which is the opposite of my claiming I've uncovered a scandal. Do you understand?

If you want to follow this stuff, read the people who explain it well , like Dylan Patel:

TSMC’s 3nm Conundrum, Does It Even Make Sense? – N3 & N3E Process Technology & Cost Detailed

Shrinking finally costs more, Moore’s Law is now dead in economic termsA couple of weeks ago, we were able to attend IEDM, where TSMC presented many details about their N3B and N3E, 3nm class…

www.semianalysis.com

I've long followed semianalysis, and indeed had the Patel/Ahmad article in mind when I wrote my first post on this thread, since they themselves carefully distinguish between new process technologies (like 3 nm vs 5 nm), and optimizations of existing process technologies (like optical shrinks on 5 nm), and explictly explains that N4P is the latter, not the former:

"...what TSMC’s N5 family consists of: N5, N5P, N5A, N4, N4P and N4X....N4 is another process optimization of N5, but it comes with a small design shrink. This is also referred to as a 'nodelet.'..In late 2021, TSMC announced N4P, a process optimization of N4."

Thus if you agree with Patel and Ahmad, you agree with me: The distinction between new process technologies and optimizations of existing process technologies is important, and we shouldn't be conflating one with the other. But that's what calling N4P "4 nm" does, since it's standard industry nomenclature to reserve new "nm" designations for new process technologies.

So it made no sense for you to get so exercised ("Oh Jesus Christ...which means fsckall" etc.) over what I wrote, particularly since you yourself are big on getting terminology right (recall your [misguided] issue with "power consumption"). Thus we both know the reason you’re reacting so strongly to my post has nothing to do with the post itself. Instead it’s because you’re still wiping the egg off your face for these two interactions, and since you don’t have an adequate response there, you’ve decided to come after me here:

https://forums.macrumors.com/threads/m2-max-geekbench-scores-surface-again-online-ahead-of-release-in-2023.2372999/page-4?post=31794239#post-31794239

https://forums.macrumors.com/threads/m2-max-geekbench-scores-surface-again-online-ahead-of-release-in-2023.2372999/page-6?post=31797562#post-31797562

 

[bradman83]

Remind me, can the 3090 be used in a laptop? Is it energy efficient? Didn't think so. Different products for different uses.

They were talking about the M1 Ultra, which also cannot be used in a laptop. The original poster was comparing the best available Apple Silicon graphics for a Mac desktop workstation with the best (consumer-grade) GPU for a Windows desktop workstation. Energy efficiency is of lesser importance to desktop users who need power more than anything.

 

[darngooddesign]

Well it sucks to be an iPhone 14 Pro owner. USB-C and 3nm for the iPhone 15 Pro. 👌

Boy... how will I ever survive not having USB-C.

Battery benefits of a 3nm iPhone are as of yet unverified so its not anything to get excited over yet.

 

[binarysmurf]

M2 pro in mini is not going to happen.
It’d kill the studio right away.

There is space in the Mac mini lineup for a machine that slots between the base M1 Mini and the Studio Max. An M2 Pro-based Mini fits the bill here and will eliminate the Intel Mini still in the line-up. This makes sense to me.

In Australia, The M1 Mini starts at $AU1099, and the Studio Max starts at $AU3099. There's a place in the lineup for the M2 Pro Mini starting at around $AU2000.

 

[MacMikePro]

I don't think anyone who actively avoids buying Gen 1 anything (cars, computers, phones, TVs) want Apple to skip the M2 Pro/Max for the MBPs (and minis and iMacs?) and wait for the M3 Pro/Max a year from now?

And, I don't think Apple wants to wait another year to take those people's money, either?

Apple isn't going to let it's best selling Macs sit at Gen 1 much longer... The Air has the M2, hell an iPad has the fraking M2... they need the MBPs to get updated sooner than the M3 Pro/Max will be available in quantity ...

That's not what I'm saying, I'm still suggesting that Apple updates the MacBook Pros in a few months, but if the chips are indeed 3nm and not based on the M2 then Apple should just call the chips M3 Pro/Max. Because if they call the chips M2 Pro/Max then that's gonna imply that they are, well, the M2 chip but more powerful and that wouldn't be the case if they are built with a different process. I definitely don't think Apple's gonna hold off on updating the MacBook Pros for another year, not at all. I'm only talking about what they name the chip, that's all.

 

[MacMikePro]

Apple is free to name anything they want any way they want. It is all marketing. No one here knows what Apple is going to do.

I agree that it is all marketing, but I still feel like there should be consistency to keep things simple. People already act like the iPad made some huge leap frog jump in performance because Apple put the M1 chip inside it when all that is is a marketing name, the M1 is literally just an A14X chip, it's just a name. It is what the iPad was always gonna get no matter what, and people just don't get that.

 

[chucker23n1]

- AMD and Intel have reached 2200 in Geekbench single-core (while Apple’s newest chip just hit 1880).

This is true, but they need far more power to achieve it.

-Qualcomm has reached 1550 in single-core

Yes: a future CPU can reach the score of an Apple CPU from two generations ago.

- Apple lost many of their best chip designers

Tell me again why we should see Apple as a leader?

Because as far as efficiency is concerned, they’re leading.

 

[Boil]

Hoping that Spring 2023 brings a Mac mini with a M2 Pro SoC on a 3nm process...

 

[Allen_Wentz]

The Digitimes says M2 Pro and Max will be 3nm in Aug 2022, this can be correct if Apple decides to delay the 14/16 Macbook releasing until Oct 2023. That will be 2 years update since M1 Pro.

Otherwise, it will be 4nm as same as A16. Obviously, we all want 3nm soon.

Not really. I want the M2 generation for MBP and Studio - - but could care less what node it comes from.

The coming Mac Pro will need 3nm for the transistor count and the bragging rights, but none of the other Apple Macs need the very latest node. All they need is 2023 engineering/architecture.

 

[Allen_Wentz]

Hoping that Spring 2023 brings a Mac mini with a M2 Pro SoC on a 3nm process...

Why? The Mini does not need the latest fab node, which by definition has lower yields and hence costs more. The entry level Minis need the most cost-effective SoC, which is very unlikely to be 3nm.

 

[svish]

Great to hear! Looking to see the new Macs that will be powered by these chips.

 

[DailySlow]

Apple's main chip supplier TSMC will kick off mass production of 3nm chips this week, with Apple being the primary customer of the new process, which could first be used in upcoming M2 Pro chips expected to power updated MacBook Pro and Mac mini models.

According to the new report by DigiTimes, TSMC will start mass production of its next-generation 3nm chip process on Thursday, December 29, in line with reports from earlier in the year that said 3nm mass production would begin later in 2022. From the report:
Apple currently uses TSMC's 4nm process in the A16 Bionic chip in the iPhone 14 Pro series but could jump to 3nm as soon as early next year. A report in August claimed the upcoming M2 Pro chips would be the first to be based on the 3nm process. The M2 Pro chip is expected to debut first in updated 14-inch and 16-inch MacBook Pros early next year and possibly updated Mac Studio and Mac mini models.

Later in 2023, according to another report, the third generation of Apple silicon, the M3 chip, and the A17 Bionic for the iPhone 15, will be based on TSMC's enhanced 3nm process, which has yet to be made available. According to the DigiTimes report today, citing industry sources, production of 3nm process chips is "unlikely to ramp up" until production of the enhanced version gets underway.

Article Link: 3nm Chips From Apple Supplier TSMC to Enter Mass Production This Week

Then there’s this…https://www.politico.com/news/2022/12/26/china-trade-tech-00072232 - could 3NM lithography/development actually be going so well, that some obfuscation could be in order as US steps back from free globalization/tech sharing for mutual development policies that prevailed (naively?) that shifted massive production dominance into China over last 30 years? #security #IP

 

[Allen_Wentz]

There is space in the Mac mini lineup for a machine that slots between the base M1 Mini and the Studio Max. An M2 Pro-based Mini fits the bill here and will eliminate the Intel Mini still in the line-up. This makes sense to me.

In Australia, The M1 Mini starts at $AU1099, and the Studio Max starts at $AU3099. There's a place in the lineup for the M2 Pro Mini starting at around $AU2000.

Agreed something needs to fill that slot. I could be an M1 Mini Pro or an M2 Mini Pro or a Studio Pro or an intel Mini or some combination thereof. Definitely an interesting product slot for Apple to fill.

 

[Allen_Wentz]

They were talking about the M1 Ultra, which also cannot be used in a laptop. The original poster was comparing the best available Apple Silicon graphics for a Mac desktop workstation with the best (consumer-grade) GPU for a Windows desktop workstation. Energy efficiency is of lesser importance to desktop users who need power more than anything.

Yes energy efficiency is of lesser importance to desktop users but energy efficiency is of some significant importance to desktop users. The dead silence of a Studio running heavy graphics is a huge value add.

 

[budda5150]

Remind me, can the 3090 be used in a laptop? Is it energy efficient? Didn't think so. Different products for different uses.

The new mobile 4090 supposedly should have ~3090 desktop performance in a laptop with TGP 150-175W. These won't likely be the thinnest machines, but the same is true if you try to shove the M1 Ultra into a laptop. I'm actually more excited about mid-range mobile Lovelace GPUs. Something like 4070 mobile should have a quite good efficiency.

 

[Boil]

The coming Mac Pro will need 3nm for the transistor count and the bragging rights, but none of the other Apple Macs need the very latest node. All they need is 2023 engineering/architecture.

Why? The Mini does not need the latest fab node, which by definition has lower yields and hence costs more. The entry level Minis need the most cost-effective SoC, which is very unlikely to be 3nm.

I was gonna argue that the Mn Max should be on the latest node because it is most likely the building block SoC for the Ultra & Extreme SoCs...

But the lower yields and higher costs thing makes sense; so maybe M2 Pro & M2 Max are on 5nm (4NP), for use in the 14" & 16" MBPs, Mac minis (Pro), & Mac Studio (Max and therefore Ultra); with the initial 3nm dies being reserved for the ASi Mac Pro, a higher cost machine to offset the higher cost of the 3nm process and to "clear the pipe" for M3 on 3nm...?

 

[black_knight]

Well it sucks to be an iPhone 14 Pro owner. USB-C and 3nm for the iPhone 15 Pro. 👌

Nah, not really. I could sell my 14 Pro for a great price since it will only be ~1 year old and buy the 15. Easy enough.