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's 2028 iPhones to Use 1.4nm A22 Pro Chips
- Thread starter MacRumors
- Start date
- Sort by reaction score
TSMC N2 this year is GAA but not BS-PDN. Next year should be BSPDN.
There are two interesting elements here that may make interop between Intel and TSMC less simple than interop between Samsung and TSMC.
1. If you have a backside metal layer, why limit it just to power delivery? Why not use it for other routine clutter (most obviously clock), so that you have more flexibility and space for signal wiring? This is feasible with how TSMC is doing BS-PDN; it is much less feasible with how Intel is doing it. (This is essentially why Intel got there first; they chose an implementation that is simpler but much less flexible and forward looking.)
This fact is of some interest to Apple, who have already submitted some patents describing how, for example, they would restructure SRAM to be faster and denser if they had access to a more sophisticated wiring layer on the back side (eg 2022 https://patents.google.com/patent/US20230298996A1 Backside Routing Implementation in SRAM Arrays)
So the question is: will Intel restructure their BSPDN to give Apple the flexibility they want? And will this happen before TSMC provide BSPDN?
2. That's fairly near term (maybe two or three years out). Slightly further out is a different idea. Your basic chip consists of a layer of logic, then multiple (20 or so) layers of metal wiring above that logic. Can you use that space, those layers of metal wiring interleaved with dielectric, for anything else? This is already being done today with MIM (Metal Insulator Metal) capacitors being placed between some metal layers, but can we be more ambitious?
Apple have suggested at least three types of functionality that could be placed in such a layer
- simplest is power transistors that control the voltage delivered to some region of the chip
- next up is the buffer/inverter/amplifiers that boost a signal that is traveling long distance from one region of the chip to another
- 3rd and fanciest is SRAM (not necessarily as dense or fast as "real" SRAM, but heck, there's a lot of area available if you can get this idea to work...)
Here's one example of such a patent: (2026) https://patents.google.com/patent/US20260090359A1 BEOL Power Switch Devices
So these all seem like good ideas. Why weren't they implemented 50 years ago? The problem is that transistor processing is hot enough to ruin metal layers, which is why we have one layer of transistors laid down, then lots of of metal layers but no more transistors.
What has changed recently is the ability to manipulate so-called 2D materials, which can provide transistor functionality, but can be manipulated at lower temperatures. Making it feasible (even if these 2D material transistors are not as performant as good old doped silicon) to lay down additional layers of transistor between metal layers.
Both Intel and TSMC claimed breakthroughs in 2D material processing in 2025. So now we see which of them is first to provide something like I've described in an actual process...
Point is, there are interesting things happening that could substantially modify how a chip is designed/fabricated, enough so that it would not be trivial (maybe not even possible) to move a chip exploiting back-side clock and signal, or top metal layer power transistors to a process that does not offer this functionality.
BTW you may have noticed a common theme in both these points – the adding functionality by multiple layers on what used to be a simple transistors followed by 20 or so metal layers. The generic "marketing" name for this CMOS 2.0 Right now the only place you'll see this name is a few publications from imec, but expect to see a lot more of it over the next decade or so. This is, in a sense, the West's equivalent to Huawei's "folded logic". It began earlier, and the first product will likely ship later, but it's vastly more ambitious and more powerful.
Pico would be better IMO. 900 picometers sounds better than 9 Angstroms.
Right on 0.5nm you probably can play Fortnite all day and the battery 🔋 won’t budge.
With those low nanometer numbers high efficiency I’m expecting all day battery 🔋 with anything you throw at it 📱
That's my Plan. My 12 Pro is still doing surprisingly well, but I think this fall is the time to pull the trigger regardless. If I wasn't worried about things continuing to get harder to get and more expensive, I'd hold off another year. But I think I want to restart my support timer before we get too much farther along.
Reason077
macrumors 601
This would be a confusing way to write it. So you’re going have the A24 chip fabricated on A14? Nah…
This has been repeat ad nauseam whenever sub-1nm chips are being discussed.
Regardless of technical accuracies there appears no better metric to describe and die process improvement but this.
ProbablyDylan
macrumors 68040
We spent how many years in the 3nm node, and how many on 5nm before that, and TSMC will be moving everything off the 2nm node after only 18 months? I have my doubts. some of those doubts are my understanding of what a nanometer node really is at this point
The next measurement down would be picometers. I remember in the 68k/ppc days when chips were measured in micrometers. Then we switched to nanometers as they got smaller. We'll switch to picometers when they drop below the 1 nanometer size.
Edit: It looks like the switch to nm happened at the end of the G4 era, when it went from 0.13 μm (130 nm) to 90 nm.
fourthtunz
macrumors 68000
Sorry I'm out of the loop, why does the future look bleak? tariffs, shortages both?
Any/all of the above, plus the fact that TSMC is inconveniently located in Taiwan (with all that entails.)
First, Musk doesn’t have a board of directors he has to convince how get a return on investment when he decides to spend 20 billion. Second, Apple’s investors are pretty confident the only real reason have $3T of market capitalization because they don’t have any expensive manufacturing plants/labor forces. They get other manufacturers to make everything for them when they want because they can leverage the scale of potential contracts for iPhones that sell about billion components a year. Apple can also point to their failed experiments where they tried to do manufacturing at scale with their own money, like growing saphire front iPhone glass panels.
IIGS User
macrumors 68000
As far as having 2 different manufacturers for chips. No single Apple device will use chips from different manufacturers. They're not dumb enough to try that stunt again.
I suspect the high end devices like the iPhone PM, Ultra, and MBP will all get TSMC produced chips. Even with the same design chips coming from different fabs are going to have different performance characteristics. Apple doesn't need some sort of processor-gate.
But the demand is sufficient enough for lower end devices to seek out a second manufacturer both to satisfy demand and have a backup strategy.
So devices like the Mac Neo, Iphone "e" whatever it is at the given time, and other non "pro" devices will get the Intel chip. These are devices that aren't exactly drawing eyeballs from consumers when it comes to benchmark speeds. Typically the people that buy these are value oriented, want "in" on the Apple ecosystem but don't have the resources or need to buy a "Pro" device. Anything going to a grade or middle school. Anything used in retail, or in a non processing intensive business environment. Health care environments. Places where data collection and tracking are more important than raw computing power.
TSMC will be used in anything "Pro". As soon as any of those hit the street someone is running a Geekbench on it, seeing how fast it is, what it can do, etc.
If I purchased a brand new iPhone XX PM or iPhone Ultra and it had the "slower" chip in it, I would be pissed. Most ProSumers would be. You pay extra money to have the latest/greatest.
If Apple is dumb enough to do that, they may as well brand it the K-Mart phone and call it a day.
I suspect the high end devices like the iPhone PM, Ultra, and MBP will all get TSMC produced chips. Even with the same design chips coming from different fabs are going to have different performance characteristics. Apple doesn't need some sort of processor-gate.
But the demand is sufficient enough for lower end devices to seek out a second manufacturer both to satisfy demand and have a backup strategy.
So devices like the Mac Neo, Iphone "e" whatever it is at the given time, and other non "pro" devices will get the Intel chip. These are devices that aren't exactly drawing eyeballs from consumers when it comes to benchmark speeds. Typically the people that buy these are value oriented, want "in" on the Apple ecosystem but don't have the resources or need to buy a "Pro" device. Anything going to a grade or middle school. Anything used in retail, or in a non processing intensive business environment. Health care environments. Places where data collection and tracking are more important than raw computing power.
TSMC will be used in anything "Pro". As soon as any of those hit the street someone is running a Geekbench on it, seeing how fast it is, what it can do, etc.
If I purchased a brand new iPhone XX PM or iPhone Ultra and it had the "slower" chip in it, I would be pissed. Most ProSumers would be. You pay extra money to have the latest/greatest.
If Apple is dumb enough to do that, they may as well brand it the K-Mart phone and call it a day.
melgross
macrumors 6502
nobody is saying that all the chip manufacture would be split, as they did when trying TSMC out against Samsung. Intel would first make the base chips while TSMC would continue to make the higher end versions. That makes sense. Later on, if Intel works out, that could change, but that would be a few years off.