Discussion Leading Edge Foundry Node advances (TSMC, Samsung Foundry, Intel)

[DisEnchantment]

TSMC's N7 EUV is now in its second year of production and N5 is contributing to revenue for TSMC this quarter. N3 is scheduled for 2022 and I believe they have a good chance to reach that target.


N7 performance is more or less understood.


This year and next year TSMC is mainly increasing capacity to meet demands.

TSMC Radically Boosts CapEx to Expand Production Capacities, To Reach $14B For 2019

www.anandtech.com

For Samsung the nodes are basically the same from 7LPP to 4 LPE, they just add incremental scaling boosters while the bulk of the tech is the same.

Samsung is already shipping 7LPP and will ship 6LPP in H2. Hopefully they fix any issues if at all.
They have two more intermediate nodes in between before going to 3GAE, most likely 5LPE will ship next year but for 4LPE it will probably be back to back with 3GAA since 3GAA is a parallel development with 7LPP enhancements.

Samsung’s Aggressive EUV Plans: 6nm Production in H2, 5nm & 4nm On Track

www.anandtech.com

Samsung's 3GAA will go for HVM in 2022 most likely, similar timeframe to TSMC's N3.
There are major differences in how the transistor will be fabricated due to the GAA but density for sure Samsung will be behind N3.
But there might be advantages for Samsung with regards to power and performance, so it may be better suited for some applications.
But for now we don't know how much of this is true and we can only rely on the marketing material.

This year there should be a lot more available wafers due to lack of demand from Smartphone vendors and increased capacity from TSMC and Samsung.
Lots of SoCs which dont need to be top end will be fabbed with N7 or 7LPP/6LPP instead of N5, so there will be lots of wafers around.

Most of the current 7nm designs are far from the advertized density from TSMC and Samsung. There is still potential for density increase compared to currently shipping products.
N5 is going to be the leading foundry node for the next couple of years.

For a lot of fabless companies out there, the processes and capacity available are quite good.

 

[SiliconFly]

...the subtitle equates 10A to 1 nm. 10A is meant to look like 10 Å (Angstroms), but nothing about 10A is actually 1 nm.

Same applies true for TSMC as well. Nothing about N2 is actually 2nm. And nothing about N3 or N5 is actually 3nm or 5nm respectively. Likewise, there is nothing 7nm in TSMC N7. It's all marketing names.

Truth is, it's been more than 2 decades since node names actually corresponded to actual physical transistor metrics like gate length, pitch, etc. Doesn't hold true for TSMC, Samsung, Intel or GF anymore.

What matters is how far ahead Intel 1nm node is compared to existing nodes like TSMC 2nm. I'd say pretty much.

 

[Aapje]

Exactly. That's also why these roadmaps are very hard to judge. What is the actual benefit we'll see from Intel 4/3/20/18/14/10 and TSMC N3E/N3P/N2/N2P/etc?

 

[igor_kavinski]

It may seem that Intel has the edge in process since they can hit 6.2 GHz on their binned KS samples but that could also be because they dedicated a lot of die space with additional transistors to enable that possibility. Zen 4 ported to Intel 7+++ may not be able to hit 6.2 GHz without sacrificing additional die space. Intel also couldn't figure out in time how to hit high frequences on Intel 4 so they canned MTL-S and gave us Raptor Lake Refresh instead. So it's kinda moot to discuss which process is best. The final product quality and performance is what matters in the end.

 

[Aapje]

Intel also still uses a monolith, which means that they have more die space for heat dissipation of the cores. They may have to sacrifice some clock speeds if they go to chiplets. This may be one of the reasons why the didn't bring Meteor Lake to desktop, although that is just speculation.

 

[jpiniero]

This may be one of the reasons why the didn't bring Meteor Lake to desktop, although that is just speculation.

I assume it's a supply/yields/cost thing. Raptor Lake is going to be the vast majority of mobile volume in 24, and seemingly in 25 too.

 

[SiliconFly]

I assume it's a supply/yields/cost thing. Raptor Lake is going to be the vast majority of mobile volume in 24, and seemingly in 25 too.

Agree. Looking at the way Intel is handling MTL supply, I think there's a good chance ARL might also experience similar supply issues. In which case, some form of Raptor Lake-Refresh-Refresh might end up being the volume product. Yikes!

 

[dullard]

I assume it's a supply/yields/cost thing. Raptor Lake is going to be the vast majority of mobile volume in 24, and seemingly in 25 too.

I assume it is a capacity thing. It isn't that Intel can't "figure out" how to make a desktop chip on Intel 4. It is that they can't make a desktop chip on Intel 4 as there just isn't the capacity. Years ago they underbought the equipment needed to make it today.

 

[igor_kavinski]

View attachment 94680

Oh no. Stinky old boring Intel 7 products beyond 2028!

 

[trivik12]

Intel 7 will be for base die or other stuff down the line. I dont think they will launch new products unless its something niche like automotive or elsewhere where mature nodes are preferred.

 

[igor_kavinski]

Samsung to introduce backside power delivery with 2nm-class production node: Report

Rumored to begin development next year.

www.tomshardware.com

Samsung trying their best to get "competitive". Hope it works out for them. It would only mean better and cheaper chips for us.

 

[SiliconFly]

Oh no. Stinky old boring Intel 7 products beyond 2028!

Raptor refresh-refresh-refresh-refresh-refresh-refresh-...

 

[SiliconFly]

Intel 7 will be for base die or other stuff down the line. I dont think they will launch new products unless its something niche like automotive or elsewhere where mature nodes are preferred.

Looking at the way they're having trouble hitting volume with their latest nodes, I wouldn't be so confident. Going forward they might choose to use 7 & 4 only for base dies. But that can happen only if the other nodes hit volume.

 

[FlameTail]

Raptor refresh-refresh-refresh-refresh-refresh-refresh-...

In our world, the raptors died when the meteor arrived, 65 million years ago.

However in the world of semiconductors, The Raptors are refusing to die, despite the Meteor 's arrival already!!

 

[Aapje]

I assume it is a capacity thing. It isn't that Intel can't "figure out" how to make a desktop chip on Intel 4. It is that they can't make a desktop chip on Intel 4 as there just isn't the capacity. Years ago they underbought the equipment needed to make it today.

I don't believe it was the plan to re-release the exact same chips with slight overclocks as 14th gen, because it makes no sense for that to have been the plan. If they knew that Meteor Lake could not be fabbed in enough volume on Intel 4 to enable a desktop release, they would have designed a new chip on Intel 7 instead of re-releasing Raptor Lake.

The old roadmaps also clearly show Meteor Lake as the client architecture, not next to Raptor Lake. I don't believe for a second that Intel ever intends for these architectures to be limited to just mobile. That is not what their roadmaps say, so I believe that this is just a way to somewhat salvage a failed architecture.

An Intel VP also publicly stated that Raptor Lake was actually added to the roadmap to bridge the gap to the delayed Meteor Lake. This makes no sense if Meteor Lake was never intended for desktop in the first place, because then the delay of Meteor Lake wouldn't have impacted the desktop.

And their 1H 2023 roadmap states that Arrow Lake on 20A is already going to be production ready by 1H 2024. If you look at the production volume chart, I see Intel 20A production expanding at the exact moment their earlier roadmap said that Arrow Lake would become ready. Coincidence?

So I think that their plan is to start producing sale-ready desktop Arrow Lakes from halfway 2024 and then to actually start selling them at the end of 2024. Perhaps even only at the higher end, keeping Raptor Lake for the lower end or perhaps even making lower end Arrow Lakes on Intel 4.

 

[jpiniero]

Plan could have been Arrow Lake on N3 and decided it was just too expensive then.

With server sales in the dumpster, they have plenty of spare 10 nm capacity too.

 

[maddie]

Intel 7 will be for base die or other stuff down the line. I dont think they will launch new products unless its something niche like automotive or elsewhere where mature nodes are preferred.

One would think this, but as you can see, Intel claims that "mature" is even larger than 14nm, and seemingly does not exist now. Can someone explain?

 

[Khato]

Let's see, what other products are soon to be released that use Intel 4/3 again? Clearly Intel just decided to use their limited Intel 4/3 production capacity on Sierra Forest and Granite Rapids instead of Meteor Lake desktop. It certainly is disappointing how dependent Intel is going to be on TSMC for a period of time due to the blatant mismanagement of the prior two CEOs.

 

[jpiniero]

It certainly is disappointing how dependent Intel is going to be on TSMC for a period of time due to the blatant mismanagement of the prior two CEOs.

That's because they were hedging their bets as to whether to dump the fabs.

 

[Khato]

That's because they were hedging their bets as to whether to dump the fabs.

Yup, the blatant mismanagement of the prior two CEOs.

 

[DrMrLordX]

And I personally take offense that the subtitle equates 10A to 1 nm. 10A is meant to look like 10 Å (Angstroms), but nothing about 10A is actually 1 nm.

And nothing about Intel 7 is 7nm, while nothing about Intel 4 is 4nm, but article writers keep making those mistakes anyway. Then there's TSMC 12nm (ha ha).

edit: @SiliconFly beat me to it. Sort of.

@jpiniero

Intel may yet dump their fabs.

@maddie

WRT your inquiry about "mature" nodes, it may be a reference to 22nm, or maybe some of the "relaxed" 14nm derivatives like 22FFL.

 

[DavidC1]

Mature nodes with IFS means other customers can use it.

Also they use to phase out their older nodes, and they won't anymore. They used to put those nodes on TSMC's older process, and they won't need to either. That's the whole point of the Tower cooperation, and now UMC.

It certainly is disappointing how dependent Intel is going to be on TSMC for a period of time due to the blatant mismanagement of the prior two CEOs.

This also has to do with the plummeting semi sales and cutting out product lines plus the awkwardness of trying to skip processes to get there faster.

For CEOs, Swan wasn't terrible, he was there for damage control to stabilize the company before Gelsinger took over. But he wasn't a visionary either. I'd say the two terrible CEOs were Otellini and Kraznich.

 

[jpiniero]

Yup, the blatant mismanagement of the prior two CEOs.

Well... Intel's products would be better if they were fabbed entirely at TSMC. And at the end of the day, Intel's a product company.

 

[DavidC1]

Well... Intel's products would be better if they were fabbed entirely at TSMC. And at the end of the day, Intel's a product company.

Disagree. Fab missteps exposed design team problems.

Even the venerable Pentium 4 "Northwood" wouldn't have succeeded if it wasn't for the process.

Pentium M, Core 2, and Sandy Bridge wouldn't have looked so impressive without their process team.

 

[SiliconFly]

Kraznich.

When it comes to totally screwing the company, Kraznich & Bob Swan hold the crown! World's top two visionaries!

 

[Aapje]

Intel obviously has design team problems as well, with Sapphire Rapids requiring 12 steppings to get right.