Intel and Apple First to Adopt TSMC's 3nm Technology

[Lightwreather]

Summary

Apple and Intel will be the first to adopt Taiwan Semiconductor Manufacturing Co.'s (TSMC) N3 (3nm) fabrication process when the contract maker of chips deploys in late 2022, Nikkei Asia news agency reported on Friday

 

Quotes

Quote

Intel is projected to use the technology to make CPUs for PCs and servers, whereas Apple is expected to use the node for its system-on-chips aimed at client devices.  

Apple and Intel are currently 'testing their chip designs' produced using TSMC's N3 process, according to a Nikkei report that cites several people with knowledge of the matter. Given that N3 is about to 'officially' enter risk production mode in the coming weeks and assuming that the information from the news agency is correct, we can speculate that Apple and Intel have already finalized their N3 CPUs and SoCs. Though it is unclear whether they have functional silicon. Commercial production of these chips is set to start in the second half of 2022.  

Intel is said to be preparing at least two products made using TSMC's N3 node: one aimed at notebooks, another for servers. There are no details about these processors at present, but Intel has already accidentally confirmed TSMC-made Xeon SoCs aimed at various niche markets. Intel confirmed months ago that it is working with TSMC on its 2023 products, but refrained from revealing any details.
Earlier this year Intel promised that its codenamed Meteor Lake and Granite Rapids processors for client PCs and high-end servers set to be made using its own 7 nm fabrication technology were due to be launched in 2023.  

The report also says that Apple is set to use its N3 SoC for its iPad tablets due in late 2022 or early 2023. Keeping in mind that Apple's latest iPad Pros use the same SoCs as the company's entry-level and midrange Macs, there is the possibility that the initial N3 SoC might also find its way into PCs. Subsequent N3-based devices from Apple will be used for smartphones.Apple did not comment on the story. TSMC's N3 manufacturing process is the company's all-new node that will provide a full node PPA (performance, power, area) improvement over N5. The foundry promises a 10% to 15% performance gain (at the same power and transistor count), up to 30% power reduction (at the same clocks and complexity), up to 70% logic density gain, and an up to 20% SRAM density gain. N3 will aggressively use extreme ultraviolet lithography (EUVL) for 'over 20 layers,' ASML once said. 

 

My thoughts

 Well, this is interesting. Not the apple part of course, that was always expected, but the Intel part. Intel, using one of it's competitor's services (competitor in terms of their fabs). It will be really interesting seeing what intel will be doing with this. Hopefully, this pushes Intel to become even more competitive. This is probably one of the more interesting and exciting times to live in the world of tech.

Sources

Tom's hardware

Nikki asia

[Gamer4714]

Intel would get the tsmc chip in their hands and try to copy it for their next gen cpus. And after that most probably return to 14nm

[Commodus]

1 minute ago, J-from-Nucleon said:

Summary

Apple and Intel will be the first to adopt Taiwan Semiconductor Manufacturing Co.'s (TSMC) N3 (3nm) fabrication process when the contract maker of chips deploys in late 2022, Nikkei Asia news agency reported on Friday

 

Quotes

 

My thoughts

 Well, this is interesting. Not the apple part of course, that was always expected, but the Intel part. Intel, using one of it's competitor's services (competitor in terms of their fabs). It will be really interesting seeing what intel will be doing with this. Hopefully, this pushes Intel to become even more competitive. This is probably one of the more interesting and exciting times to live in the world of tech.

Sources

Tom's hardware

Nikki asia

Let's see whether Intel can make good use of the process and deliver products on time that compete well against Apple and AMD chips. This is the company that decided it would rather be stuck on 14nm for years rather than admit it needs help, after all.

 

It's good news if Intel can finally leap to a sub-10nm process. The problem, of course, is that Apple and AMD aren't standing still.

 

 

[LAwLz]

TSMC N3 should be a massive jump for Intel.

Hell, even TSMC 5nm would be a massive jump from Intel's 10nm SuperFin, and that's not even the process Intel uses for their desktop products.

 

TSMC's transistor density for their 3nm node is expected to be slightly below 300MTr/mm².

For comparsion, Intel's 10nm SuperFin process sits at slightly above 100MT/mm².

 

Imagine being able to almost triple the amount of transistors per square mm.

 

The future is looking bright for Intel, and as a result us consumers too.

 

 

  

8 minutes ago, Gamer4714 said:

Intel would get the tsmc chip in their hands and try to copy it for their next gen cpus. And after that most probably return to 14nm.

That's not how it works...

[Gamer4714]

1 minute ago, LAwLz said:

TSMC N3 should be a massive jump for Intel.

Hell, even TSMC 5nm would be a massive jump from Intel's 10nm SuperFin, and that's not even the process Intel uses for their desktop products.

 

TSMC's transistor density for their 3nm node is expected to be slightly below 300MTr/mm².

For comparsion, Intel's 10nm SuperFin process sits at slightly above 100MT/mm².

 

Imagine being able to almost triple the amount of transistors per square mm.

 

The future is looking bright for Intel, and as a result us consumers too.

 

 

  

That's not how it works...

Ok, I guess I don't know much about these

[porina]

11 minutes ago, Gamer4714 said:

Intel would get the tsmc chip in their hands and try to copy it for their next gen cpus. And after that most probably return to 14nm.

What is there to copy? Intel copying an Intel design? They have all the info already.

 

It you're suggesting copying the fab methods, there's enough secret sauce that looking at it isn't going to help much. Buying a cooked meal doesn't tell you exactly how it was prepared. Ingredients are relatively easy, but how do you put them together?

 

10 minutes ago, Commodus said:

This is the company that decided it would rather be stuck on 14nm for years rather than admit it needs help, after all.

Who exactly could they have gone to for help? This is leading edge stuff. Pretty much only TSMC, Intel and Samsung are playing the leading edge game. Others have given up. and are sticking to older processes.

 

 

On the Intel side, it is an interesting question exactly what they will use TSMC 3nm for. I wouldn't expect consumer CPUs to be in the 1st wave. Server maybe, or GPU related products. Don't forget they're going tile big time, so it isn't even necessarily a whole product, but one part of such.

 

[cj09beira]

24 minutes ago, LAwLz said:

TSMC N3 should be a massive jump for Intel.

Hell, even TSMC 5nm would be a massive jump from Intel's 10nm SuperFin, and that's not even the process Intel uses for their desktop products.

 

TSMC's transistor density for their 3nm node is expected to be slightly below 300MTr/mm².

For comparsion, Intel's 10nm SuperFin process sits at slightly above 100MT/mm².

 

Imagine being able to almost triple the amount of transistors per square mm.

 

The future is looking bright for Intel, and as a result us consumers too.

 

 

  

That's not how it works...

problem is it wont be for consumer products most likely, tsmc doesn't have the capacity for it, it will likely be for lower volume products, maybe their new gpus, fpgas etc

[BuckGup]

Better late than never but Intel has a lot of catching up to do still and AMD isn’t sitting around like Intel did

[pas008]

38 minutes ago, BuckGup said:

Better late than never but Intel has a lot of catching up to do still and AMD isn’t sitting around like Intel did

i'm not sure they are that far behind considering their 14nm+ to the billionth is still very competitive with amds 7nm

but yes amd isnt sitting around either so amds lead could get very dominant if intel does keep dragging their feet

[sazrocks]

Interesting; I wonder where AMD is at in this. Apple is going to be taking the lion's share of capacity, and once Intel has had their presumably also large slice of the pie, how much capacity will be left for AMD?

[Forbidden Wafer]

1 minute ago, sazrocks said:

Interesting; I wonder where AMD is at in this. Apple is going to be taking the lion's share of capacity, and once Intel has had their presumably also large slice of the pie, how much capacity will be left for AMD?

AMD supposedly is taking Apple's share on 5nm as they move to 3nm.

[NeuesTestament]

25 minutes ago, Forbidden Wafer said:

AMD supposedly is taking Apple's share on 5nm as they move to 3nm.

Wouldn't that be problematic longterm for AMD if they aren't able to get bleeding edge manufacturing? Doesn't sound like something you want to have happen.

[StDragon]

3 hours ago, Gamer4714 said:

Intel would get the tsmc chip in their hands and try to copy it for their next gen cpus. And after that most probably return to 14nm

That's not how that works.

 

CPU architecture design is predicated on what node they will be using. More transistors per unit of area opens up many opportunities for optimizations; such as signaling, core counts, cache, instruction sets, and overall more transistors to increase the IPC rating.

 

It's a lot like building a house; the design would be optimal for the amount of square footage you have to work with. Or in the case of a CPU, shrink the material so you can have more specialized rooms for the same given area.

[cj09beira]

3 hours ago, pas008 said:

i'm not sure they are that far behind considering their 14nm+ to the billionth is still very competitive with amds 7nm

but yes amd isnt sitting around either so amds lead could get very dominant if intel does keep dragging their feet

it really isn't competitive, look at power usage and die size and it becomes clear what the disadvantages are, intel is forced to use much more silicon to be able to compete, its simply that frequency is no longer scaling with each node so clocks dont change much.

3 hours ago, NeuesTestament said:

Wouldn't that be problematic longterm for AMD if they aren't able to get bleeding edge manufacturing? Doesn't sound like something you want to have happen.

amd is using tiny dies these days, they would have no issue using a new node that still has yielding issues, but why fight against apple for wafer allocation when you can use 5nm still provide a great upgrade for the market and let apple suffer through the teething issues of the node.

i really doubt intel is going to release cpus on 3nm anytime soon, the node is not really ready for it, i am guessing whatever they have bought is for some other product 

[pas008]

58 minutes ago, cj09beira said:

it really isn't competitive, look at power usage and die size and it becomes clear what the disadvantages are, intel is forced to use much more silicon to be able to compete, its simply that frequency is no longer scaling with each node so clocks dont change much.

if intel was on tsmc's 7nm node, wouldnt they would be the clear winner across the board?

and thats not even including chiplets yet

[sounds]

Most likely Intel paid more than AMD, so Intel gets 3nm chiplets and AMD gets 5nm chiplets.

 

Everyone will be using chiplets in 2022. That much is already public.

[Kisai]

8 hours ago, Gamer4714 said:

Intel would get the tsmc chip in their hands and try to copy it for their next gen cpus. And after that most probably return to 14nm

Definitely not how it works.

 

What's likely going to happen is that Intel will temporarily use TSMC for the laptop dies (since that's where they are now lagging behind, and uncompetitive with AMD), Intel currently uses the laptop dies (Coffee Lake/Ice Lake) in the E/W/Platinum Xeon's, and also hasn't released any E/D Xeon's in 2 years. 

 

I don't see Intel doing consumer CPU's on TSMC's N3 process as those are not the highest in demand/most profitable part, unless they decide to use the same core on everything by moving to chiplets.

 

[cj09beira]

4 hours ago, pas008 said:

if intel was on tsmc's 7nm node, wouldnt they would be the clear winner across the board?

and thats not even including chiplets yet

not really comparing to current design they would gain perf/w but most likely loose quite a bit of max frequency as tsmc's node isn't as focused on max performance, so no gain performance wise, but they could go for more cores, maybe 10, even still at most it might have forced amd to slightly lower the cost of their cpus, but overall amd would still have a better group of cards to play with 

[CarlBar]

12 hours ago, Gamer4714 said:

Intel would get the tsmc chip in their hands and try to copy it for their next gen cpus. And after that most probably return to 14nm

 

Everyone, pretty sure this is suppressed to be a joke.

 

Also as someone said, TSMC doesn't have the volume for intels needs, they are not moving the bulk of their silicon to 3nm, TSMC just can't produce enough. This is either a specialised component of a chiplet setup, or a specialised overall product.

[Commodus]

On 7/2/2021 at 10:08 AM, porina said:

Who exactly could they have gone to for help? This is leading edge stuff. Pretty much only TSMC, Intel and Samsung are playing the leading edge game. Others have given up. and are sticking to older processes.

 

 

On the Intel side, it is an interesting question exactly what they will use TSMC 3nm for. I wouldn't expect consumer CPUs to be in the 1st wave. Server maybe, or GPU related products. Don't forget they're going tile big time, so it isn't even necessarily a whole product, but one part of such.

They could've gone to TSMC for help sooner, I figure.

 

And you're right on initial models... there was talk Intel might use TSMC for laptop chips, but even that might be a stretch. Apple's coziness with TSMC makes it much more likely to secure orders for mainstream computer chips. It's funny to think that Intel (and even a bit of the Windows monopoly) could be undone by manufacturing capacity limits.

[WolframaticAlpha]

Good to see Intel being less arrogant these days.

[AzzaNezz]

Explain it to me what does that mean.

Since Intel is at the moment on 10 nm,and AMD 7nm,but still Intel is somewhat competing with AMD,does that means when Intel goes to 3nm they will be far ahead on AMD?

 

[LAwLz]

1 hour ago, AzzaNezz said:

Explain it to me what does that mean.

Since Intel is at the moment on 10 nm,and AMD 7nm,but still Intel is somewhat competing with AMD,does that means when Intel goes to 3nm they will be far ahead on AMD?

It's a big complicated but there are some things that are important to know and understand.

1) There are multiple foundries that make chips. The big, leading edge ones are TSMC, Intel and Samsung.

2) Intel has historically made their own chips at their own foundries. AMD and Apple uses TSMC for their current chips. Nvidia uses Samsung for their current 30 series. These things quite often change though. For example AMD does not have to use TSMC if they don't want to. Right now they do want to however because TSMC have the best manufacturing process.

3) The nanometer numbers you see get thrown around can not be compared between foundries. TSMC 5nm is not the same as Intel 5nm, nor is TSMC 8nm the same as Samsung's 8nm.

4) There is more to processor design than just the manufacturing process (3nm, 10nm, etc).

5) Even within the same foundry and same "nm size", there can be differences. You've probably heard of Intel putting a bunch of pluses behind their nodes (like 14nm++), a lot of foundries does that (although it's memed far more with Intel because they have had to do it more than others). For example TSMC has N7FF and N7FF+. The latter allows for almost 20% more transistors in the same area.

 

A far more important metric than nanometer is transistor density. That is to say, how many transistors are you able to pack into a given area (usually measured in millions of transistors per square mm). This is a far easier and better metric to use when trying to compare process nodes from different foundries.

 

Here are the transistor density (millions of transistors per square mm) for some different manufacturing nodes. Higher is better although please note that even using the same manufacturing process, transistor density numbers can vary. Theses are just estimates and guidelines:

 

• Intel 14nm++ - 37.2
• Intel 14nm - 37.5
• TSMC 10nm - 52.51
• Samsung 8nm - 61.2
• Samsung 7nm (second gen) - 95.3
• TSMC 7nm - 96.5
• Intel 10nm - 100.8
• Samsung 7nm (third gen) - 112.5
• TSMC 7nm+ - 113.9
• TSMC 6nm - 114.2
• Samsung 5nm - 127
• TSMC 5nm - 173
• Intel 7nm - 202-250
• TSMC 3nm - 291

 

Intel is currently competitive with AMD because their chips are larger and uses more power than AMD's. With a more advanced process node, Intel will be able to reduce power consumption as well as increase performance.

Intel on 3nm would be far ahead of AMD's current processors that are using 7nm. But it's not like Intel will release 3nm processors tomorrow, and AMD will not sit idle either.

What might happen is that if Intel uses TSMC 3nm for their processors (which has not been confirmed, it can be for other things) then Intel might be ahead of AMD for a while, until AMD also gets on a comparable process node (maybe also 3nm TSMC?), and then it will be down to who got the best microarchitecture.

[AzzaNezz]

15 minutes ago, LAwLz said:

It's a big complicated but there are some things that are important to know and understand.

1) There are multiple foundries that make chips. The big, leading edge ones are TSMC, Intel and Samsung.

2) Intel has historically made their own chips at their own foundries. AMD and Apple uses TSMC for their current chips. Nvidia uses Samsung for their current 30 series. These things quite often change though. For example AMD does not have to use TSMC if they don't want to. Right now they do want to however because TSMC have the best manufacturing process.

3) The nanometer numbers you see get thrown around can not be compared between foundries. TSMC 5nm is not the same as Intel 5nm, nor is TSMC 8nm the same as Samsung's 8nm.

4) There is more to processor design than just the manufacturing process (3nm, 10nm, etc).

5) Even within the same foundry and same "nm size", there can be differences. You've probably heard of Intel putting a bunch of pluses behind their nodes (like 14nm++), a lot of foundries does that (although it's memed far more with Intel because they have had to do it more than others). For example TSMC has N7FF and N7FF+. The latter allows for almost 20% more transistors in the same area.

 

A far more important metric than nanometer is transistor density. That is to say, how many transistors are you able to pack into a given area (usually measured in millions of transistors per square mm). This is a far easier and better metric to use when trying to compare process nodes from different foundries.

 

Here are the transistor density (millions of transistors per square mm) for some different manufacturing nodes. Higher is better although please note that even using the same manufacturing process, transistor density numbers can vary. Theses are just estimates and guidelines:

 

• Intel 14nm++ - 37.2
• Intel 14nm - 37.5
• TSMC 10nm - 52.51
• Samsung 8nm - 61.2
• Samsung 7nm (second gen) - 95.3
• TSMC 7nm - 96.5
• Intel 10nm - 100.8
• Samsung 7nm (third gen) - 112.5
• TSMC 7nm+ - 113.9
• TSMC 6nm - 114.2
• Samsung 5nm - 127
• TSMC 5nm - 173
• Intel 7nm - 202-250
• TSMC 3nm - 291

 

Intel is currently competitive with AMD because their chips are larger and uses more power than AMD's. With a more advanced process node, Intel will be able to reduce power consumption as well as increase performance.

Intel on 3nm would be far ahead of AMD's current processors that are using 7nm. But it's not like Intel will release 3nm processors tomorrow, and AMD will not sit idle either.

What might happen is that if Intel uses TSMC 3nm for their processors (which has not been confirmed, it can be for other things) then Intel might be ahead of AMD for a while, until AMD also gets on a comparable process node (maybe also 3nm TSMC?), and then it will be down to who got the best microarchitecture.

Great thanks for explaining,this is was really informative.

[leadeater]

1 hour ago, LAwLz said:

A far more important metric than nanometer is transistor density. That is to say, how many transistors are you able to pack into a given area (usually measured in millions of transistors per square mm). This is a far easier and better metric to use when trying to compare process nodes from different foundries.

@AzzaNezzAnother additional factor to consider for different foundries and node processes is voltage frequency curves. Traditional Intel processes have been designed to target higher power and higher voltages than TSMC/Samsung/GlobalFoundries as their primary customer base is lower power, mobile and lower frequency (GPU) customer designs. This is how Intel gets those frequency and overclocking advantages as the node (and architecture design) has a less aggressive voltage increase as frequency increases at the upper limits.

 

Two node processes with the same transistor densities will not actually produce the same results, one could achieve lower power at moderate frequencies but suffer a steep voltage and power increase ramp as frequency increases and the other could have worse power efficiency at moderate frequencies however be able to achieve high peak frequencies at lower voltages. It's the balance between power efficiency and performance, some nodes target performance more than power efficiency i.e. Intel 14nm but that's not to say this process wasn't considered power efficient when it was first introduced but the design goal was performance.