TSMC's Next-Generation Chip Technology for Apple Silicon on Schedule

[Populus]

I think I’ll finally pull the trigger on the M4 gen if it includes significant changes on the architecture, a beefier Neural Engine ready for the upcoming AI capabilities, and hopefully more base RAM. I think it will be a good time to replace my 2014 Mac mini.

As for the iPad Pro, it’s been rumored (thanks @vadimyuryev) that it will also come with an M4 so that will definitely be interesting to see…

And why next one, and not the current gen? Well, I suspect the changes in the A18/M4 generation are going to be bigger than previous generations, as the A17/M3 gen was just a clock boost taking advantage of the smaller node. With the couple additions of AV1 decoding support and Ray Tracing.

And why the next iteration of Apple Silicon and not keep waiting? Well, from what I’m reading, TSMCs 2nm will come in 2026, that’s a long time to keep waiting, and it could happen that they got bad yields just like happened with the first version of their 3nm node.

I’d like to jump to M4 with the next Mac mini, along with an M4 iPad Pro and an A18 iPhone. The problem is, that future iPad Pro will come with an OLED display and I don’t know if my eyes will suffer just like with some OLED iPhone (PWM). And the upcoming iPhone SE is not only coming with the cheapest OLED screen, but also will be significantly bigger. So I’m afraid I’ll be stuck with my SE 3 for a long time.

 

[MayaUser]

I think I’ll finally pull the trigger on the M4 gen if it includes significant changes on the architecture, a beefier Neural Engine ready for the upcoming AI capabilities, and hopefully more base RAM. I think it will be a good time to replace my 2014 Mac mini.

As for the iPad Pro, it’s been rumored (thanks @vadimyuryev) that it will also come with an M4 so that will definitely be interesting to see…

And why next one, and not the current gen ? Well, I suspect the changes in the A18/M4 generation are going to be bigger than previous generations, as the A17/M3 gen was just a clock boost taking advantage of the smaller node. With the couple additions of AV1 decoding support and Ray Tracing.

And why the next iteration of Apple Silicon and not keep waiting? Well, from what I’m reading, the 2nm TSMC will come in 2026, that’s a long time to keep waiting, and it could happen that they got bad yields just like happened with the first version of their 3nm node.

ipad pro will not come this year with M4, but M3
I think only Mac Pro or Mac Studio can , maybe "small chances" at WWDC with M4..but until now always the iPhone set the trend with the new base SoC

 

[Populus]

ipad pro will not come this year with M4, but M3
I think only Mac Pro or Mac Studio can , maybe "small chances" at WWDC with M4..but until now always the iPhone set the trend with the new base SoC

There’s a source that points towards the next iPad Pro jumping directly to M4, and if they are unveiling M4 gen during WWDC, and we don’t get the new iPad Pros (that keep being delayed) until summer, then they will probably come with the M4 SoC.

I know it sounds crazy, but remember that iPad Pro gets refreshed every 1.5 years, and if they release the M3 iPad Pro in may, and during the following month they announce the M4 Apple Silicon family, the flagship iPad will be almost all of it’s lifespan with an outdated SoC.

I’m not saying it will happen, but it could.

Source:

Also it is rumored that the new N3E node has been in production since December 2023…

 

[Rigby]

angstroms. Intel is calling their "2nm" process 20A, and they are allegedly coming out with chips using 20A *this year*. If they actually do that, it will put them back in the process lead for the first time in about a decade.

Both TSMC "2nm" and Intel "20A" will bring a significant change compared to recent steps: they will both switch to GAA transistors with these generations. Intel will also introduce backside power delivery with 20A for the first time. They are also bringing a bunch of very interesting packaging technologies to the market. Interesting times ahead.

 

[Populus]

Both TSMC "2nm" and Intel "20A" will bring a significant change compared to recent steps: they will both switch to GAA transistors with these generations. Intel will also introduce backside power delivery with 20A for the first time. Interesting times ahead.

True. That’s the only reason that might tempt me to wait a bit longer. But 2 more years seem a bit too much already (at least for me with my old Intel Mac mini).

 

[MayaUser]

There’s a source that points towards the next iPad Pro jumping directly to M4, and if they are unveiling M4 gen during WWDC, and we don’t get the new iPad Pros (that keep being delayed) until summer, then they will probably come with the M4 SoC.

I know it sounds crazy, but remember that iPad Pro gets refreshed every 1.5 years, and if they release the M3 iPad Pro in may, and during the following month they announce the M4 Apple Silicon family, the flagship iPad will be almost all of it’s lifespan with an outdated SoC.

I’m not saying it will happen, but it could.

Source:

View attachment 2367339
Also it is rumored that the new N3E node has been in production since December 2023…

dont have high hopes...why?
1) Apple i bet they still have enough M3 SoC...around, Apple is about profit margins
2) Apple we know sometimes can update the same product even twice in 12 months so M3 now, and M4 spring 2025
3) iPhones comes first...like every time
4) When you want to release a brand new SoC, if you dont do it with your most popular product then do it with the most powerful and thats Mac Pro

5) We heard these before with M3...M3 will come before iphone etc etc...and it didnt

 

[Populus]

dont have high hopes...why?
1) Apple i bet they still have enough M3 SoC...around, Apple is about profit margins
2) Apple we know sometimes can update the same product even twice in 12 months so M3 now, and M4 spring 2025
3) iPhones comes first...like every time
4) When you want to release a brand new SoC, if you dont do it with your most popular product then do it with the most powerful and thats Mac Pro

5) We heard these before with M3...M3 will come before iphone etc etc...and it didnt

Yeah, I don’t have high hopes either, your points are valid.

However, maybe this year the M4 family is presented at WWDC, and the A18 with the iPhone 16 a couple of months later.

 

[deconstruct60]

....
Apple chipmaker TSMC is making progress toward manufacturing 2nm and 1.4nm chips that are likely destined for future generations of Apple silicon, DigiTimes reports.

....​

The manufacturing time frames for mass production of 2nm and 1.4nm chips have now apparently been determined: Trial production of the 2nm node will begin at in the second half of 2024, with small-scale production ramping up in the second quarter of 2025.

Digitimes also said that TSMC N3 was planned to release in late 2022 also.

https://www.macrumors.com/2021/12/23/tsmc-3nm-chip-production-4q22/


TSMC is going to be their first node with "Gate all around" which they have not done at volume scale before. There is decent chance they will hit a hiccup in their plan. That is in part why the bounds are loose like "second half" as opposed to "Quarter n" .

'small scale' ramping late 2Q '25 is a bit late for the iPhone chip. (That's May-June and iPhone needs to ramp output in April-May. Even more so if the bake time ( even more multipatterning than N3 ) is longer. If the 'bake' times are getting longer then the gap between trail and high volume is likely to get longer also. The process bug-fix cycle gets longer, so the QA improve cycle is likely going to take longer. ( can try to offset that with faster wafer cycling time through newer EUV machines, but adding more complexity/patterns also. Those also effectively eat up cycle times. ) . TSMC's target for N2 is "second half '25" which is relatively late for long bake times that have to finish by June-August to have time to build large inventory for September.

• 
  
• iPhone 15 Pro (2023): A17 Pro (3nm, N3B)
• iPhone 16 Pro (2024): "A18" (3nm, N3E)
• "iPhone 17 Pro" (2025): "A19" (2nm, N2)
• "iPhone 18 Pro" (2026): "A20" (2nm, N2P)
• "iPhone 19 Pro" (2027): "A21" (1.4nm, A14)

Decent chance Apple does not skip N3P to hit the 2025 target and the above fab node usage all shifts down a year to next generation.

N3P is more aligned with what Apple is doing with their silicon than N3E is. If they spend a substantial amount of money re-spinning from N3B to N3E ... why throw a chunk of that out the window as quickly as possible? N3P is design rule compatible with what they just did with N3E while N2 is not.

TSMC's 'second half Y202_ ' volume production rollouts have fallen into a misalignment with the iPhone "every September , technically arbitrary launch requirement.

 

[Nozgog]

Yes N3P is missing - hopefully it is used this year instead of N3E - otherwise if N2 has delay that means N3P in 2025

 

[macduke]

Notably, TSMC's new plant in Arizona will also join 2nm production efforts.

I think this is the most important piece of news in this entire story. I've seen many experts assume that since they are setting up these plants for 3nm that they were locked into that, but at the rates TSMC moves, that just didn't make sense. I figured they would all be upgraded as time progresses with the plant in Taiwan, with perhaps a small lag. I get the sense that this thinking comes from American factories being setup in a more permanent way while these overseas factories are probably much more efficient and modular. We need to adopt this sort of agile factory system if we want to actually make America great at manufacturing again.

 

[Darren.h]

Bring on 2nm Mac Studios with 196GB max ram and Thunderbolt 5!! I want One.

 

[NT1440]

I think this is the most important piece of news in this entire story. I've seen many experts assume that since they are setting up these plants for 3nm that they were locked into that, but at the rates TSMC moves, that just didn't make sense. I figured they would all be upgraded as time progresses with the plant in Taiwan, with perhaps a small lag. I get the sense that this thinking comes from American factories being setup in a more permanent way while these overseas factories are probably much more efficient and modular. We need to adopt this sort of agile factory system if we want to actually make America great at manufacturing again.

I don’t understand where that Arizona factory is going to get all the water these fabs require, unless this is long term planning as all the elderly who live there start to die off…

 

[Rigby]

I think this is the most important piece of news in this entire story. I've seen many experts assume that since they are setting up these plants for 3nm that they were locked into that, but at the rates TSMC moves, that just didn't make sense. I figured they would all be upgraded as time progresses with the plant in Taiwan, with perhaps a small lag. I get the sense that this thinking comes from American factories being setup in a more permanent way while these overseas factories are probably much more efficient and modular. We need to adopt this sort of agile factory system if we want to actually make America great at manufacturing again.

It should be mentioned that 2nm production in Arizona is scheduled to begin in 2028, years after their Taiwan fabs. It's not likely that the US fabs will ever use their most advanced processes at any point in time.

 

[bodonnell202]

I'll be surprised if the A19 is actually produced on N2 as N3P seems like a more likely candidate, but time will tell...

 

[ChrisA]

So the size is going from 12 to 4 to 3 to 1.2 and so on. It will not be long until we are at zero nanometers and then what? Obviously, there will be negative nanometers. First, we will have -0.1nm and then -2.0, and later in the century maybe "minus 30".

Something like this happened before. The ancient Greeks did not seem to mind much that the year was 525 BC and that the next year would be 524 BC. But the Romans because very nervous starting at round 5BC. They wanted to know what would happen five years after "zero". Some said the world would end. But it turned out to be very uneventful and they just started to count upward instead of the backward counting they had been doing for thousands of years.

The flip to negative nanometers will be different. chips use less and less power as the features get smaller but negative size features actually generate power. So 100 years from now we will plug in our phones not to charge the phone but to create power to run the lights and refrigerator.

Or I could be wrong and the numbers are meaningless, We will just have to wait and see.

 

[eoblaed]

Gonna run out of nm soon

 

[bodonnell202]

So the size is going from 12 to 4 to 3 to 1.2 and so on. It will not be long until we are at zero nanometers and then what? Obviously, there will be negative nanometers. First, we will have -0.1nm and then -2.0, and later in the century maybe "minus 30".

Something like this happened before. The ancient Greeks did not seem to mind much that the year was 525 BC and that the next year would be 524 BC. But the Romans because very nervous starting at round 5BC. They wanted to know what would happen five years after "zero". Some said the world would end. But it turned out to be very uneventful and they just started to count upward instead of the backward counting they had been doing for thousands of years.

The flip to negative nanometers will be different. chips use less and less power as the features get smaller but negative size features actually generate power. So 100 years from now we will plug in our phones not to charge the phone but to create power to run the lights and refrigerator.

Or I could be wrong and the numbers are meaningless, We will just have to wait and see.

I'm assuming this is a joke, bit in case it isn't...

This actually has happened before when semiconductor process technology was generally measured in μm, and once they reached 1 μm they switched to using nm, so they could move the decimal 3 places (i.e. after 1 μm came 800 nm, which is 0.8 μm). As we are approaching 1 nm it seems like the preference is to use Angstroms (Å), which only moves the decimal once place, so 1 nm = 10 Å, which is why TSMC plans to call their process after 2nm A14 instead of N1.4.

 

[mrhick01]

TSMC has the best fabs in the world.

Apple should encourage and finance the construction of at least one cutting-edge fab (not the fab that makes two—or three-gen old console processors) in the United States, even if that fab costs, say, $40 billion for the construction, labor, and personnel training.

It's expensive, sure, but it would be a hedge against...future predicaments

 

[Westside guy]

If a speedometer measures speed, then a nanometer measures nan…. Grandmas? Flatbread?

I think that would be a naanometer.

 

[steve09090]

There is no industry-wide agreement among different manufacturers about what numbers would define a "3 nm" node

Who put the USB Implementers Forum, Inc in charge? Were they bored, and when will they take over the putting in of speed signs on highways?

 

[steve09090]

I think that would be a naanometer.

I like naan, and usually for the cheese version. Yummy.

 

[Tagbert]

If I understood it correctly the "x nm" is just arbitrary and not an actual size? What comes after?

not exactly arbitrary, but more like relative sizing.

 

[svish]

Good to know about this. Wonder when M4 will be released.

 

[smallduck]

When is TSMC due to figure out backside power delivery too? This tech might end up being a big deal, equivalent to halving the process size again.

 

[bradman83]

In the US "nanometer" (nm) may seem arbitrary for most people because it is metric. Nano is 10^-9, ergo move the decimal point 9 spaces to the left (e.g. 1 nm = 1 x 10^-9 = 0.000000001 metres) 1 metre ~ 3' - 3.37"

1 nm ~ 0.00000003937 inches

No one’s saying its arbitrary because it’s metric, they’re saying it’s arbitrary because it has no bearing on the actual size of the transistor components within a chip.

A 3nm process nodes means that the process could theoretically create components 3nm in size, but in practice the transistors inside of a chip are much larger than that.