The entire market took a major hit because of the pandemic. It has only just recently recovered.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Apple to Get TSMC's First 2-Nanometer Chips
- Thread starter MacRumors
- Start date
- Sort by reaction score
The entire market took a major hit because of the pandemic. It has only just recently recovered.
From the horse’s mouth:
How we get to 1 nanometer chips and beyond
At this year’s IEDM conference, we lay the groundwork for a near future where semiconductors with nodes at 1nm and smaller are possible.
And:
Though we’re well beyond the 18 month Moore’s law situation.
Please see https://en.wikipedia.org/wiki/2_nm_process, second paragraph, and the table below. “2 nm” is really just around 10% smaller than “3 nm”.
The SRAM/cache and I/O "transistors" for N3E are the same size at N5. Everything isn't shrinking at the same rate. It isn't 'smaller' transistors as much as it 3D density of transistors. The 'nm' stuff is mostly marketing handwaving at this point.
Mainly because Intel screwed up their usage of N3B. Mediatek beat Apple to usage of N4. Nvidia is far, far , far more flush with cash now. They don't 'have to' wait.
N3P hasn't shipped in volume yet. There are likely some test chips in at risk production at this point, but not really in the "already shipping" status. It is due 2H 2024
[News] TSMC’s N3P to Enter Mass Production in the Second Half of Next Year, Rumored Inclusion of Tesla in Customer List | TrendForce News
TSMC, the world's leading foundry in the 3nm technology, is reportedly experiencing a surge in the number of New Tape-Outs (NTOs) for the 3nm family i...
www.trendforce.com
Which paints a picture whether Apple uses N3P or some late 2H 2025 process in 2025. Apple being the primary consumer of N2 isn't a sure thing that this Digitimes spins makes it out to be. 2H 2025 would be a large 'miss' for the A-series ( which needs to ramp in March-April to be avaialble in sufficient numbers by September in finished devices. .). N2 is likely to need just as long a 'bake time' as N3 has. It is unlikely to be early (ahead of schedule) in 2025.
After nanometer comes picometer. So maybe we'll go to 800pm process?
Because performance gains will become insignificant, the focus will switch from hardware to improving software, fixing the bugs, making the user interface intuitive again …
Picometer... then femtometer.
Intro to Size and Scale | Exploring the Nanoworld
University of Wisconsin Materials Research Science and Engineering Center Interdisciplinary Education Group
chem.beloit.edu
Considering the machines you are using now (and them being good enough), then none of these improvements are likely to matter to you. And same for me. My 2018 mini was "good enough" though sometimes lagged just a touch. My M2 Pro mini is more power than I can use (setting aside that I tried gaming on it for the first time on a Mac in a decade, that is my one use case that pushes my M2 Pro mini).
Even when the lambda parameter was still measured in hundreds of microns, it was never about the "actual physical size of the transistors", it was about the scale of the design rules. Typically, the minimum feature size would have been 2 lambda and transistors might be around 10 lambda.
As processes push harder on diffraction and other limits, design rules can no longer be as independent of scale as they once were, but it has no less to do with "the actual physical size of the transistors or anything" than it ever did.
Eventually when silicon transistors get too small, electrons start moving in seemingly unpredictable ways. You have to move past silicon to something like carbon nanotubes.
Angstroms.
1nm = 10 Angstroms, so reducing in single Angstrom increments is around 1/10th the improvement of a 1nm jump.
There's really not that great of an improvement to be made anymore.
The focus will shift more to cameras and other features - storage, cooling, battery tech.
Does this translates in savings or speed improvments for customers?. Is my 15PM opening apps faster than my "old" 13PM or faster than a competitor phone?. Will this restrict features to scale down to older phones (I am seeing you smart HDR only possible for some obscure reasons on the 10% faster CPU)....
At about 1.5nm to 1nm Moore’s Law dies, some have said [not sure if anyone has mentioned this yet]. Transistors are made of silicon atoms arranged in a crystal lattice the smallest physical size is .2nm so about .5nm using silicon.
Last edited:
Savings ? Pretty likely not . N3 costs substantively more than N7 and N5 . N2 is even more complex to make with even more multipatterning . it is not getting cheaper .
TSMC's average wafer price jumped 22% in one year — nearly all semiconductor industry growth now comes from more expensive products, not higher production volumes
Shipments of 3nm wafers account for 15% of the company's revenues.
The side effect of Moore’s law that generally lead to incrementally lower prices is mostly over. It is at point we’re asking for speed and savings is a bit of a conflict. Pick one or the other.
there is a notion of ‘cheaper’ if users are moving down the product line. Used to “have to “ buy a Mac Pro but now Can buy a Mini Pro. Butin the same product class , that is probably not going to move . There was lots of hype “arm” was going to dramatically drive down system costs. Not going to happen.
Naw, we will start getting negative nm chips that are infinitely fast and actually produce power while computing. 🤓
No, it didn't. It came from the 1 extra core added over the 5 existing ones.
I'm so excited for 2nm chips to become mainstream; I just KNOW Safari's going to be extra-snappy when I leave goofy comments on MacRumors articles. I might need to put on some goggles just so I can handle all this speed.
Such a simple viewpoint.
You have NO IDEA how awesome Sugar Smash will be with all that computing power.
I’m all ready calculating tips in calculator on my 15PM at LightSpeed.
It is a given that Apple will get priority to any die shrink.
Apple will be the first company to receive chips built on the TSMC's future 2-nanometer process, DigiTimes said today. According to sources that spoke to the site, Apple is "widely believed to be the initial client to utilize the process."
The report comes from DigiTimes "Tomorrow's Headlines" alert, so additional details may be available in the full news story.
TSMC is expected to begin producing 2nm chips starting in the second half of 2025. Terms like "3nm" and "2nm" refer to the specific architecture and design rules TSMC is using for a family of chips. Decreases in node size correspond to a smaller transistor size, so more transistors can fit on a processor, leading to boosts in speed and more efficient power consumption.
This year, Apple adopted 3-nanometer chips for its iPhones and Macs. Both the A17 Pro chip in the iPhone 15 Pro models and the M3 series chips in Macs are built on the 3-nanometer node, an upgrade over the prior 5nm node. The jump from 5nm technology to 3nm technology brought notable 20 percent faster GPU speeds, 10 percent faster CPU speed, and a 2x faster Neural Engine to the iPhone, and similar improvements on Macs.
TSMC is building two new facilities to accommodate 2nm chip production, and working on approval for a third. TSMC generally builds new fabs when it needs to increase production capacity to handle significant orders for chips, and TSMC is expanding in a major way for 2nm technology. The transition to 2nm will see TSMC adopting GAAFET (gate-all-around field-effect transistors) with nanosheets instead of FinFET, so the manufacturing process will be more complex. GAAFETs allow for faster speeds with a smaller transistor size and lower operational voltage.
TSMC is spending billions on the change, and Apple will also need to make chip design changes to accommodate the new technology. Apple is TSMC's main client, and it is typically the first to get TSMC's new chips. Apple acquired all of TSMC's 3-nanometer chips in 2023 for iPhones, iPads, and Macs, for example.
In between the 3nm and 2nm nodes, TSMC will introduce several new 3nm improvements. TSMC has already come out with N3E and N3P chips that are enhanced 3nm processes, and there are other chips in the works such as N3X for high performance computing and N3AE for automotive applications.
Rumors suggest that TSMC is already starting work on more advanced 1.4-nanometer chips, which are expected to come out as soon as 2027. Apple is said to be looking to reserve TSMC's initial manufacturing capabilities for both 1.4nm and 1nm technologies.
Article Link: Apple to Get TSMC's First 2-Nanometer Chips
Decent chance AMD is using N3E
AMD Shares New CPU Core Roadmap, 3nm Zen 5 by 2024, 4th-Gen Infinity Architecture
To Infinity Architecture, and beyond
And
“…
It remains to be seen which of TSMC's 3nm-class process technology will be adopted by Nvidia. TSMC has numerous 3nm nodes, including performance-enhanced N3P
…
Nvidia will, of course, not be alone with the adoption of TSMC's N3 technology next year: AMD, Intel, MediaTek, and Qualcomm are all set to adopt one of the foundry's 3nm-class nodes in 2024 – 2025. In fact, MediaTek has already taped out its first N3E design with TSMC.
…”
Nvidia's Blackwell B100 GPU to Hit the Market with 3nm Tech in 2024: Report
Nvidia to adopt TSMC's 3nm-class process technology next year.
Many of the folks the came with N4 solutions instead of targeting N3B are queued up to go with N3E.
N3P is problematical for anyone trying to target the majority of 2024. If TSMC tuns on HVM on N3P august-September then really would have problems delivering volume in December. ( likely looking at 3+ month ‘bake‘ time. )
Recent reports are that Intel will be consume N3B which has side effect off consuming lots of resources .
Pretty decent chance Mediatek is going to get to N3E volume first.
( AMD is likely going to ship N4 zen 5 first. The N3 aimed at a narrower segment they feel they can wait longer on )
Last edited: