MacRumors

macrumors bot
Original poster
Apr 12, 2001
54,211
16,027


Apple chipmaker TSMC has set out the potential performance and power efficiency of the upcoming iPhone 12's A14 chip, according to a report by AnandTech.

tsmc_semiconductor_chip_inspection_678x452.jpg


The iPhone 12 is expected to contain an A14 chip, based on TSMC's smaller 5nm manufacturing process. The iPhone 11's A13 chip used a 7nm process by comparison.

Manufacturing chips using this smaller process offers a number of advantages, including a reduction in power consumption of up to 30 percent, and up to 15 percent increased performance. This indicates what kind of improvements may come to the A14 chip in the iPhone 12.

There is a tradeoff when selecting one or the other, but Apple tends to prioritize performance improvements over power consumption. When reducing the size of the manufacturing process for a new generation of chips, power consumption and performance improvements are slightly lower.

Last year, TSMC announced a $25 billion investment in the new 5nm node technology in a bid to remain the exclusive supplier of Apple's processors. TSMC has reportedly been in mass production of chips using the 5nm process for several months now, and the process is also expected to be used for Apple Silicon chips coming to the Mac before the end of this year.

In addition to the 5nm process for 2020, TSMC outlined its plans for a 3nm process coming in late 2022. This would likely be used for a potential A16 chip and other future Apple Silicon if the company follows previous years, but it is understandably difficult to speculate on Apple's manufacturing plans so far ahead. The 3nm process yields similar 30 percent and 15 percent power consumption and performance improvements over the 5nm process.

It is worth noting that regardless of the performance of TSMC's chips, Apple usually optimizes its software for further performance improvements. It is also possible that Apple could significantly prioritize power consumption over performance for improved battery life. These decisions belong to Apple regardless of TSMC's manufacturing process, making the exact behavior of the A14 chip somewhat uncertain until official announcement.

Article Link: TSMC Details Potential iPhone 12 A14 Performance and Upcoming 3nm Process
 
Last edited:
  • Like
Reactions: RandomDSdevel

Daniel L

macrumors 6502
Sep 15, 2009
455
151
Bring on the iPhone 12. TSMC has surpassed Intel in state of the art lithography processes. Apple must be paying a premium to have the first consumer device with chips manufactured on the 5nm node.
 
Comment

Jbusick7944

macrumors 6502
Oct 8, 2008
254
437
I don't understand this statement in the third sentence. In two other places it says smaller = faster and more efficient...so what's up with this part? "When reducing the size of the manufacturing process, power consumption and performance improvements face a slight reduction."
 
Comment

cmaier

macrumors Penryn
Jul 25, 2007
24,000
30,124
California
I don't understand this statement in the third sentence. In two other places it says smaller = faster and more efficient...so what's up with this part? "When reducing the size of the manufacturing process, power consumption and performance improvements face a slight reduction."
It’s gibberish.

It’s also not true that smaller necessarily means faster and more efficient, but that’s just me being pedantic.
 
Comment

cmaier

macrumors Penryn
Jul 25, 2007
24,000
30,124
California
We are about to hit the atomic limit of process nodes.
Nope. They told me that in 1992, too. :)

0.2 nm or so for silicon, yes? Got a ways to go. And, of course, the process node “size” has very little to do with the actual size of actual devices. I don’t think I ever used a minimum length transistor.
 
  • Like
Reactions: CarlJ
Comment

jav6454

macrumors P6
Nov 14, 2007
17,197
2,239
1 Geostationary Tower Plaza
Nope. They told me that in 1992, too. :)

0.2 nm or so for silicon, yes? Got a ways to go. And, of course, the process node “size” has very little to do with the actual size of actual devices. I don’t think I ever used a minimum length transistor.
Well, device size does not mean process node size. In this case a Silicone atom is 0.2nm in length. We are approaching that limit in which single atoms become the Transistors vs now a few atoms make a transistor.
[automerge]1598367961[/automerge]
TSMC is amazing. How come Intel can't do it?
Complacency in their position.
 
Comment

cmaier

macrumors Penryn
Jul 25, 2007
24,000
30,124
California
Well, device size does not mean process node size. In this case a Silicone atom is 0.2nm in length. We are approaching that limit in which single atoms become the Transistors vs now a few atoms make a transistor.
[automerge]1598367961[/automerge]

Complacency in their position.

Remember that the process node is not the size of the transistor - it would be the minimum length of a transistor gate. (More or less - depends on how a fab names its nodes). That assumes lateral transistors (not fins), assumes a single finger, and ignores that transistor gates also have width. And all of that also ignores that the transistor also has a source and a drain. And of course, all of that assumes not just lateral transistors, but assumes the transistors are FETS.

So the relation is pretty weak.
 
  • Like
Reactions: CarlJ
Comment

iReality85

macrumors 65816
Apr 29, 2008
1,082
2,345
Upstate NY
We are about to hit the atomic limit of process nodes.
Nope. They told me that in 1992, too. :)

0.2 nm or so for silicon, yes? Got a ways to go. And, of course, the process node “size” has very little to do with the actual size of actual devices. I don’t think I ever used a minimum length transistor.

Conceivably, if nodes keep being reduced by roughly 1/2 for the next decade, we'll enter the atomic range soon:

2019: 5nm

2022: 3nm

2025? 1nm - 1.5nm?

Then we enter picometers.

2030? 500pm - 750pm? (.5nm - .75nm)

Depending on the atom, that's anywhere from 50pm - 500pm in size. Basically, silicon's days will be numbered once we reach the 2030s.
 
Comment

cmaier

macrumors Penryn
Jul 25, 2007
24,000
30,124
California
I think it's supposed to suggest that performance and power consumption gains are becoming less pronounced at each new stage.

Shows a graph of performance and power-consumption improvements.


If so, that’s misleading. The graph is phony - the spacing on the x-axis is the same even thought the difference in quantity is different.
 
Comment

cmaier

macrumors Penryn
Jul 25, 2007
24,000
30,124
California
Conceivably, if nodes keep being reduced by roughly 1/2 for the next decade, we'll enter the atomic range soon:

2019: 5nm

2022: 3nm

2025? 1-1.5nm?

Then we enter picometers.

2030? 500pm - 750pm (.5nm - .75nm)

Depending on the atom, that's anywhere from 50pm - 500pm in size. Basically, silicon's days will be numbered once we reach the 2030s.

Maybe. But, again, the number being in the atomic range doesn’t mean much. A lot depends on the geometry of the transistor, and the actual size of its components, and the “node size” doesn’t tell you too much about that.

Plus we’re really in the silicon-germanium era right now (using germanium to create kinks in the conduction/valence bands to increase carrier lifetime), and switching to a different material (like GaAs/AlGaAs) doesn’t help much if what you are worried about is not being able to shrink any more.
 
  • Like
Reactions: Coolkiwi and CarlJ
Comment
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.