Discussion Intel current and future Lakes & Rapids thread

[JasonLD]

The future appears quite clear for anyone without bias.

2023 production for nodes using EUV
TSMC = 370 Kwpm
Intel = 20 Kwpm

Wouldn't that be enough for whatever Intel has planned for 2023? Since logic chiplets for Meteor Lake is going to be the only thing Intel will be using their 7nm.

 

[Saylick]

Wouldn't that be enough for whatever Intel has planned for 2023? Since logic chiplets for Meteor Lake is going to be the only thing Intel will be using their 7nm.

Possibly, Mizuho Securities shows Intel 7nm being risk production all the way out to the end of 2022, so maybe it would be ready for HVM in 2023. Also, whether or not 20 kwpm is adequate depends on how much of their line-up they need to shift over to Meteorlake to be competitive with AMD by 2023, who will most likely be at the tail end of Zen 4, with Zen 5 on TSMC N3 likely coming in H1 2024. N3 looks to ramp faster than N5, which itself ramped faster than N7, so Apple could potentially move onto N3 faster than normal, leaving a glut of N5P volume for AMD in 2023. If AMD is pushing out 50 kwpm of Zen 4 in 2023, 20 kwpm for Meteorlake may not be sufficient.

 

[IntelUser2000]

Don't count Intel out. Surprises happen because nobody expects them.

There were rumors Intel had absolutely terrible times and yields on the 22nm process but things improved quite drastically in a year or so. Yes, since 22nm came 6 months later than previous nodes that rumor is plausible.

Also, don't treat "Intel" and "AMD" like you treat "David" and "Bob" and "Nancy". Corporations are not individuals. Sure, individuals form a group to run them. Things can change.

We tend to go from one extremes to the other. Same with investment. Ethereum actually dropped to low as $130 last year before rocketing to nearly $5000 this year.

As long as it doesn't die it recovers. And I mean die as in actually dying not "die" as in it dropped so low that vast majority ignores it.

 

[bsp2020]

Don't count Intel out. Surprises happen because nobody expects them.

It takes months to install EUV machines and get it operational. There is no way Intel will have enough EUV machines to produce their main products in volume in 2023. We will hear about Intel buying up EUV machines at least 1, if not 2 years before Intel can ship large volumes of 7nm products.

https://seekingalpha.com/article/4424007-asml-stock-in-sweet-spot (Figure 2C)

Based on that info, I doubt we will see any volume shipment of Intel 7nm products at least till 2024 at the earliest. AMD EPYC will be well ahead of Intel at least till then.

 

[LightningZ71]

Is it possible that there will be a further iteration of their 10nm process? They already have 10og, base 10, 10sf, and 10esf. There's always room for a 10esf+ or something. Their 10 isn't so bad with respect to density. As they ramp up production on the later nodes, they could just throw more transistors at the problem and keep clocks stagnant.

 

[Saylick]

Is it possible that there will be a further iteration of their 10nm process? They already have 10og, base 10, 10sf, and 10esf. There's always room for a 10esf+ or something. Their 10 isn't so bad with respect to density. As they ramp up production on the later nodes, they could just throw more transistors at the problem and keep clocks stagnant.

It's completely possible, but at some point if they don't get their 7nm node to the same levels of volume as their 14nm/10nm nodes, they can end up with a similar scenario with 10nm as with 14nm, i.e. where they ended up with multiple generations of the same node because their newest node wasn't yielding enough, leading to a node disadvantage across their product line. If their ultimate strategy is to concede the cutting edge nodes to TSMC, then maybe it's not a big deal, but that'd be a big blow to US national interests.

 

[irfacilities]

it obviously uses something similar (fair) engineering as in the customer Ice Lake, Tiger Lake, Lakefield, and Rocket Lake CPUs. All in all, this is attempted and tried (and deferred) engineering. Similarly, (or considerably more) significant for the server farm, a center check is essentially helped from 28 to 40. This is answerable for the >40% execution improvement Intel guarantees, and affirmed by surveys, and sometimes up to almost +80%.)

 

[eek2121]

It takes months to install EUV machines and get it operational. There is no way Intel will have enough EUV machines to produce their main products in volume in 2023. We will hear about Intel buying up EUV machines at least 1, if not 2 years before Intel can ship large volumes of 7nm products.

https://seekingalpha.com/article/4424007-asml-stock-in-sweet-spot (Figure 2C)

Based on that info, I doubt we will see any volume shipment of Intel 7nm products at least till 2024 at the earliest. AMD EPYC will be well ahead of Intel at least till then.

IIRC we know little about Intel’s 7nm process, including how heavily it utilizes EUV.

Is it possible that there will be a further iteration of their 10nm process? They already have 10og, base 10, 10sf, and 10esf. There's always room for a 10esf+ or something. Their 10 isn't so bad with respect to density. As they ramp up production on the later nodes, they could just throw more transistors at the problem and keep clocks stagnant.

A while ago, Intel admitted 10nm would be a short lived node. Whether they were talking about 10nm as a whole or 10nm vs. 10+, 10SF, etc. I have no idea. It seems like Intel wants to get to 7nm ASAP. Don’t be surprised if they fudge the numbers a bit to get there.

 

[jpiniero]

It's completely possible, but at some point if they don't get their 7nm node to the same levels of volume as their 14nm/10nm nodes, they can end up with a similar scenario with 10nm as with 14nm, i.e. where they ended up with multiple generations of the same node because their newest node wasn't yielding enough, leading to a node disadvantage across their product line. If their ultimate strategy is to concede the cutting edge nodes to TSMC, then maybe it's not a big deal, but that'd be a big blow to US national interests.

You are talking about a much smaller die. I'm assuming the only part of Meteor Lake that's 7 nm is just the CPU Core and it's L2 cache. Everything else is some other node. I do still wonder if that tile is even dual sourced at TSMC.

 

[DrMrLordX]

Wouldn't that be enough for whatever Intel has planned for 2023? Since logic chiplets for Meteor Lake is going to be the only thing Intel will be using their 7nm.

That means Intel will be stuck on nodes that are roughly competitive with TSMC nodes from three years ago (N7).

 

[jpiniero]

That means Intel will be stuck on nodes that are roughly competitive with TSMC nodes from three years ago (N7).

Intel's 7 is more comparable to TSMC 5, presuming that Intel hasn't gutted it.

 

[maddie]

Don't count Intel out. Surprises happen because nobody expects them.

There were rumors Intel had absolutely terrible times and yields on the 22nm process but things improved quite drastically in a year or so. Yes, since 22nm came 6 months later than previous nodes that rumor is plausible.

Also, don't treat "Intel" and "AMD" like you treat "David" and "Bob" and "Nancy". Corporations are not individuals. Sure, individuals form a group to run them. Things can change.

We tend to go from one extremes to the other. Same with investment. Ethereum actually dropped to low as $130 last year before rocketing to nearly $5000 this year.

As long as it doesn't die it recovers. And I mean die as in actually dying not "die" as in it dropped so low that vast majority ignores it.

All that you've posted here is true, but, it appears that Intel is dependent on factors outside of their control.

The present substrate shortage as a limiting factor for ramping all chip production is a very good example for us. In this case, the limiting item output can be expanded quickly, but not in the EUV case.

You can have the most perfect design with 100% yields and it will be relatively ineffective if you can't make enough at the appropriate time.

 

[DrMrLordX]

Intel's 7 is more comparable to TSMC 5, presuming that Intel hasn't gutted it.

I meant Intel 10nm and variants. @JasonLD was basically saying, Intel would be okay with low-volume 7nm in 2023 since it would only be for a few small things, to which I made my response.

 

[eek2121]

That means Intel will be stuck on nodes that are roughly competitive with TSMC nodes from three years ago (N7).

Intel 10nm is competitive with TSMC 7nm. Intel 7nm is competitive with TSMC 5nm.

 

[maddie]

Intel 10nm is competitive with TSMC 7nm. Intel 7nm is competitive with TSMC 5nm.

Exactly. 3 yrs behind for any production volume that matters.

TSMC 3nm is starting volume shipping soon.

 

[rainy]

Intel 10nm is competitive with TSMC 7nm. Intel 7nm is competitive with TSMC 5nm.

On marketing slides, definitely./s

Btw, would you name any Intel product made in 7nm (not from TSMC of course)?

 

[jpiniero]

Btw, would you name any Intel product made in 7nm (not from TSMC of course)?

This whole thread was sparked because they announced they were going to do some SiFive IP on their 7 nm.

 

[maddie]

This whole thread was sparked because they announced they were going to do some SiFive IP on their 7 nm.

This whole thread?

Intel current and future Lakes & Rapids thread

 

[jpiniero]

This whole thread?

Intel current and future Lakes & Rapids thread

Well, the last page or two.

 

[maddie]

Well, the last page or two.

I forgot I'm living in the age of hyperbole.

 

[Saylick]

IIRC we know little about Intel’s 7nm process, including how heavily it utilizes EUV.

I could be overly simplifying this, but whether it uses 1 layer of EUV vs. 10 layers of EUV doesn't change the fact that there's at least one step in the litho process where EUV is required, and therefore it can be a bottleneck if Intel doesn't have enough EUV machines.

Here's what I could quickly find on Google in comparing TSMC N5 vs. Intel 7nm:

1) TSMC N5
https://fuse.wikichip.org/news/3398/tsmc-details-5-nm/

TSMC emphasized the extensive use of EUV with this process. It’s worth pointing out that this is really TSMC’s first ‘main’ EUV-based process. TSMC N7 and N7P nodes are DUV-based. TSMC first production EUV process is the N7+ but that node is really an orphan – not compatible with the prior nodes and no clear migration path forward other than going back to this node. On the other hand, N5 is designed as the main migration path from N7 for most customers. TSMC says that more than 10 EUV layers are used to replace at least 4 times more immersion layers at cut, contact, via and metal line steps. This is comparing their EUV-based N5 node to a hypothetical N5 node that utilizes multi-patterning.

2) Intel 7nm
https://www.anandtech.com/show/13683/intel-euvenabled-7nm-process-tech-is-on-track

By contrast, Intel’s 7 nm production tech will use extreme ultraviolet lithography (EUVL) with laser wavelength of 13.5 nm for select layers, reducing use of multipatterning for certain metal layers and therefore simplifying production and shortening cycle times. As it appears, the 7 nm fabrication process had been in development separately from the 10 nm tech and by a different team. As a result, its development is well underway and is projected to enter HVM in accordance with Intel’s unannounced roadmap, the company says.

Intel has never disclosed characteristics of its 7 nm fabrication tech, but a major reduction of multi-patterning usage as well as a more traditional 2X scaling goal vs. 10 nm indicates a more extensive usage of EUVL.

According to ASML, one EUV layer requires one EUV step-and-scan system for every ~45,000 wafer starts per month. Therefore, if Intel plans to use EUVL extensively for 10 to 20 layers, it will require approximately 20 to 40 EUVL scanners for a fab with a 100,000 wafer starts per month capacity. Considering that Intel is not the only company with plans to use EUVL in the 2020s, getting the number of EUVL scanners it might need for HVM at multiple fabs may be a challenge.

So, if Mizuho's estimates are to be believed, by 2023 they will produce 20 kwpm of 7nm wafers. If we assume 45 kwpm corresponds to 1 machine per layer, and we use this chart to assume Intel has 20 EUV machines by EOY 2023, then either A) Intel uses a ton of EUV layers for 7nm (~45 layers of EUV, 20 machines per layer per 45 kwpm, so 20 kwpm if the EUV machines are running at full capacity), or B) Intel uses 10--20 layers of EUV and is struggling to get their EUV machines up to speed, or C) Mizuho's estimates are low balling the true production amount.

A) is unlikely. B) is more likely. C) is probably true.

 

[jpiniero]

It takes time to setup the EUV machines. I have no idea how long but you would have to account for that.

 

[Ajay]

Intel 10nm is competitive with TSMC 7nm. Intel 7nm is competitive with TSMC 5nm.

On what? Test sleds? [small Logic + SRAM chips]. Actual products are the real measure. Intel 10nm CPUs have decent performance, but worse perf/watt. Luckily for Intel, AMD is well behind in wafer availability [from TSMC]. CPUs built on Intel's 7N will not have the wafer availability, and hence, the same advantage over AMD [on TSMC N5, looking at the Mizuho data] The bulk of Intel's product will still have to be on 10N ESF in 2023. There is no way to magically make up for the difference without reducing the effective xtor density and efficiency (less EUV layers) and making huge dies to claw back performance.

 

[Saylick]

It takes time to setup the EUV machines. I have no idea how long but you would have to account for that.

So you're saying B) is what's happening, which if that is the case then I think it's taking Intel too long.

 

[Ajay]

It takes time to setup the EUV machines. I have no idea how long but you would have to account for that.

From a recent video on ASML I watched, about 6mo from delivery to full qualification (ready for HVM).