Apple to get TSMC's initial 2nm capacity, says obvious report

AppleInsider

A new rumor suggests that Apple has already queued up for the majority of TSMC's early 2nm production, but it is perhaps the most obvious report to come out of the supply chain yet.

M3 is built with the 3nm process

Apple saturated TSMC's 3nm production with its A17 Pro and M3 processor family, so the company no doubt has plans to do it again for the next process. The 2nm generation may not arrive for some time, however, as it's not expected to go into production until 2025.

According to a Digitimes headline flagged for release on January 25, Apple will get TSMC's initial 2nm capacity. Previous rumors suggest this would be for the iPhone 17 Pro lineup due in late 2025.

Sometimes, supply chain rumors are a little too on the nose, like a weatherman predicting warm weather in the summer. We didn't need an official report to know it was true.

We'll hear more about Apple's investment in the 2nm process as we get closer to 2025. For now, M3 still needs to reveal an Ultra tier in the 3nm process.

Rumor Score: Likely

Read on AppleInsider

9secondkox2

2025 is moving pretty quick if true. 

3nm just got here. 

A 2026 availability is just two years after 3nm. 

MisterKit

How low can you go?

daven

I’m looking forward to it but I’ll almost certainly get a 3nm MacBook this year.

danox

No surprise Apple can pay and their designs are finished while most of the competition dithers in nebulous AI....

blastdoor

9secondkox2 said:

2025 is moving pretty quick if true. 

3nm just got here. 

A 2026 availability is just two years after 3nm. 

Agreed! The historical trend is a new process every 18 to 24 months. So this is totally in line with that.

Intel has claimed they will leapfrog TSMC and come out with their 2nm process later this year. I’ll believe it when I see it, though, since intel’s track record hasn’t been very good.

blastdoor

Being first in line for new processes from TSMC gives Apple the potential to catch up and surpass the competitor in AI. 

ajmas

Though hoping improvements in Apple's chips isn't solely dependant on the 2nm process. Single core improvements will be appreciated, since not all processing can benefit from more cores.

Marvin

MisterKit said:

How low can you go?

Roughly 1nm, they scale down around 0.7x each time 7nm, 5nm, 3nm, 2nm.

https://www.tomshardware.com/tech-industry/manufacturing/tsmc-charts-a-course-to-trillion-transistor-chips-eyes-monolithic-chips-with-200-billion-transistors-built-on-1nm-node

Next ones are 1.4nm, 1nm and they are starting to use Angstroms (=0.1nm) so A14, A10.

This sites says there's research for going below 1nm:

https://www.tomshardware.com/news/imec-reveals-sub-1nm-transistor-roadmap-3d-stacked-cmos-20-plans

But, a Silicon atom is 0.2nm so they aren't planning to scale down linearly after 1nm and instead scale out in 3D with stacks. This would allow more transistors in the same footprint but there's also the issue of power draw. Adding 3 stacks can get 3x the power in the same size of chip but if it's using 3x the power then it's no good for consumer chips.

It will get extremely expensive to invest in that kind of density of chip production. 2nm = $28b.

There won't be such a big consumer demand for it when chips reach that level. More of the demand will be for servers as power vs footprint is important.

Consumer chips could easily stop at 1nm. This would be Macbook Air equivalent to M3 Max and Max chip equivalent to a desktop Nvidia 4090, around 2030.

BlueLightning

Several chip manufacturers have experimented with nanotubes over the years.  Some military production has used materials other than silicon or silicon carbide.  China responded to embargoes on chip building equipment with news that they will switch to optical chips in the future, and China restricted export of materials used for non-silicon chips, such as gallium (which has military chip applications).  Maybe we are due for a replacement of integrated circuit chips.  Multiple states per bit of storage, rather than just 0 and 1 states (too expensive in the past).  Computing has already transformed from mechanical tabulators, to relays/delay lines/plug boards in earlier times, to electronic vacuum tubes, discrete transistors, and the (mostly current) integrated circuits.  Cost could cause a prolonged pause in speed increases, similar to other technologies when they reach a certain level of development (with little change over decades of time).  It could be interesting or very boring, depending on what happens in 10-20 years.  A revisited analog computing implementation could be in the future.  I've not spent the time to understand quantum computing (or even quantum physics), and suspect it is too expensive and complex for normal home or small/medium business use.  

Suspect we may see a continuation of more cores per chip, cores dedicated for specific uses, and multi-chip implementations.  I understand some data center servers now have 8 or so GPUs for AI/ML functions, paired with 2 or so CPUs for traditional workloads.  Amazon, Microsoft, Google and other hyper-scalers have been designing their own chips and servers, in some cases for 5-10 years.  

Some chip designers have roadmaps of planned future chips.  Several have experienced longer delays between planned generations in more recent years.  For many home users, improved performance may not be needed, especially for applications like web browsing, spreadsheets, document creation and emails.