[SemiAccurate] Intel kills off the 10nm process

[stockolicious]

https://semiaccurate.com/2019/01/25/why-semiaccurate-called-10nm-wrong/

Charlie's got a new article regarding 10 nm. He agrees with me now that Intel instead of killing 10 nm off completely will just ship some relatively low volume parts and will just have to be creative with binning.

Intel IMO needs to go to 7 ASAP (ie: Products out in 1H 2021)

They have a lot to do to hit 2021 - they need a new design (Most likely an MCM and they need to tailor it to the new node) One of charlie's articles had an INTC internal memo saying they wont be competitive until 2023. I think maybe getting a CEO in place immediately would help too - who is running that place?

 

[moinmoin]

Do you think that TSMC/Samsung will share their know-how with Intel?

Why there are not companies offering 14/10/7nm in China, Russia or Middle east? They have capital to buy whatever they want. Why it is Exclusive only in Taiwan/Korea? Maybe because it is not just that simple as buying 7nm-ready tool from ASML and pressing power button.

Of course TSMC/Samsung won't share their know-how with Intel expect when licensed. Of course ASML doesn't deliver complete fabs that work on the press of a power button. But the opposite extreme, ASML delivering the tools and every single manufacturer starting EUVL R&D from scratch is equally inane. The whole EUVL research had been going for near two decades now, including essentially a deployment delay of more than one decade, it a huge feedback loop with a lot of different participating companies.

 

[DrMrLordX]

Do you think that TSMC/Samsung will share their know-how with Intel?

If Intel makes it worth their while, sure. I can see a future in which Intel, TSMC, Samsung, and many smaller fabs pool resources to reach the limits of transistor density. TSMC might make it their on their own, but the likelihood that they'll do so "in good order" would be significantly increased if they cooperated with others. I, for one, do not really favor the formation of monopolies, but from a strictly technological point-of-view, it appears to me that many established foundry companies may wash out well before reaching the limits of process shrinks.

Why there are not companies offering 14/10/7nm in China, Russia or Middle east? They have capital to buy whatever they want. Why it is Exclusive only in Taiwan/Korea? Maybe because it is not just that simple as buying 7nm-ready tool from ASML and pressing power button.

Mubadala thought they could buy their way into the semiconductor foundry business. They already threw in the towel. It's not just a matter of having the resources . . . it's about being able to monetize your product after you've poured billions of dollars into research and development. Companies like TSMC and Samsung are still profiting from their drive towards 3nm. GF gave up. Russian and Chinese concerns seem content to pick off what tech they can from pioneers so that they can push also-ran nodes at a discount. Intel's future is hazy. They still have a solid shot at 7nm, but there is a risk that they'll wash out too.

 

[Phynaz]

Since for 7nm Intel will be using EUV tools and those will be already in use for some time at TSMC and Samsung Intel may be able to avoid early pitfalls. That is if they don't deviate from common industry practices and try to bite off more than they can chew again.

Intel has had EUV tools on prem for longer than Samsung or TSMC. In 2012 Intel poured $4.2B into ASML to fund development. I believe intel was the only buyer of EUV equipment for the first year or two. I’m pulling this out of my butt, but I recall intel ordering the first four EUV machines, and it took a year for ASML to build them.

I believe intel has had to go through all the pitfalls. At the time intel was far out on the bleeding edge. This was before the 10nm debacle.

 

[Phynaz]

I would not be naive enough to believe intels anyone’s claims about process before they really deliver.

I expanded upon your statement

 

[jpiniero]

They have a lot to do to hit 2021 - they need a new design (Most likely an MCM and they need to tailor it to the new node) One of charlie's articles had an INTC internal memo saying they wont be competitive until 2023. I think maybe getting a CEO in place immediately would help too - who is running that place?

They have the MCM - after all that's what Cascade Lake-AP and Cooper Lake are; not to mention the Rapids with EMIB/Forveros. But yeah, they can't wait. If they can't deliver 7 in 2021 they need to talk to TSMC or Samsung and fab there, even if it's only the CPU tile.

Imagine if AMD gets to TSMC 5 nm while almost all of Intel's server sales are still 14 nm, and that's just AMD.

 

[Ajay]

They have the MCM - after all that's what Cascade Lake-AP and Cooper Lake are; not to mention the Rapids with EMIB/Forveros. But yeah, they can't wait. If they can't deliver 7 in 2021 they need to talk to TSMC or Samsung and fab there, even if it's only the CPU tile.

Imagine if AMD gets to TSMC 5 nm while almost all of Intel's server sales are still 14 nm, and that's just AMD.

If Intel can’t do 7nm in volume, they will lose game. No way TSMC can ramp enough volume for Intel.

 

[jpiniero]

With Apple's settlement with Qualcomm today, it is now rumored they will be using Qualcomm's 5G modem in the 2020 iPhone. That's of course real bad news for Intel's modem efforts and likely means the modem project will be shut down now as they don't have any customers. We'll probaly never know what node it was on, but if it was really 10 nm, it was doomed anyway.

 

[jpiniero]

And sure enough... Intel is exiting the modem business.

https://www.businesswire.com/news/h...Intel-Exit-5G-Smartphone-Modem-Business-Focus

 

[maddie]

If Intel can’t do 7nm in volume, they will lose game. No way TSMC can ramp enough volume for Intel.

Then we're all screwed, as by extension TSMC will not be able to fab enough for AMD to accommodate demand, as I assume that more customers will abandon Intel and shift to AMD.

TSMC can't fab for the entire market. Prices will rise.

 

[JasonLD]

Intel's modem was always inferior compared to Qualcomm's anyways. Probably better for them to focus on other areas.

 

[Ajay]

Then we're all screwed, as by extension TSMC will not be able to fab enough for AMD to accommodate demand, as I assume that more customers will abandon Intel and shift to AMD.

TSMC can't fab for the entire market. Prices will rise.

Yeah, Intel would still be filling orders because of base demand; AMD Zen2 CPUs would just become a premium product. I’m sure TSMC would be able to give AMD more wafers - just nothing close to what Intel produces. So, oddly, it’s better for consumers if Intel ships 7nm EUV products on time.

 

[ondma]

Yeah, Intel would still be filling orders because of base demand; AMD Zen2 CPUs would just become a premium product. I’m sure TSMC would be able to give AMD more wafers - just nothing close to what Intel produces. So, oddly, it’s better for consumers if Intel ships 7nm EUV products on time.

Why is it odd? Competition is good, right? That is what we heard for years when AMD was trailing.

 

[Markfw]

Why is it odd? Competition is good, right? That is what we heard for years when AMD was trailing.

Its odd, since, if AMD could provice all the product that people wanted, the monopoly could be over, and we could see real competition. 10-20% share is not real competition. They way I see it, AMD could get much higher (and even in Intels VP's own words) if they can provide enough product.

 

[ondma]

Its odd, since, if AMD could provice all the product that people wanted, the monopoly could be over, and we could see real competition. 10-20% share is not real competition. They way I see it, AMD could get much higher (and even in Intels VP's own words) if they can provide enough product.

And whose fault is it that AMD is capacity limited? (if they in fact are, I haven't seen information either way recently).

 

[dark zero]

Then we're all screwed, as by extension TSMC will not be able to fab enough for AMD to accommodate demand, as I assume that more customers will abandon Intel and shift to AMD.

TSMC can't fab for the entire market. Prices will rise.

You forgot Samsung... if TSMC can't, Samsung will

 

[NTMBK]

Then we're all screwed, as by extension TSMC will not be able to fab enough for AMD to accommodate demand, as I assume that more customers will abandon Intel and shift to AMD.

TSMC can't fab for the entire market. Prices will rise.

Good job AMD is only putting the CPU chiplet on 7nm, and keeping the northbridge on cheap'n'cheerful 14nm.

 

[DrMrLordX]

And sure enough... Intel is exiting the modem business.

Wow. No 5G modems at all? So how many products does that mean they're going to kill before launch? The 8060 was already canned last year (I think?) so . . . the 8160? I guess we don't have to guess at where they were fabbing it now. Or we could if we enjoy idle speculation.

 

[maddie]

Good job AMD is only putting the CPU chiplet on 7nm, and keeping the northbridge on cheap'n'cheerful 14nm.

That certainly helps.

Thinking about it further, the DIY PC builder might not be as badly affected as I first thought. AMD should prioritize the retail market, as the ASP is higher than the bulk deals with OEMs, especially for the higher performing SKUs, which should be the ones we're most interested in.

At least, I hope so.

 

[coercitiv]

Thinking about it further, the DIY PC builder might not be as badly affected as I first thought. AMD should prioritize the retail market, as the ASP is higher than the bulk deals with OEMs, especially for the higher performing SKUs, which should be the ones we're most interested in.

The DYI PC market is easy to deal with, they'll keep churning Ryzen 2000 at GloFo and feed the budget sensitive segment with last gen chips. As long as they sell every 7nm chip that sees the light of day, any other 12nm product that feeds the market to keep ASP in check is a win in terms of market share.

 

[jpiniero]

Thinking about it further, the DIY PC builder might not be as badly affected as I first thought. AMD should prioritize the retail market, as the ASP is higher than the bulk deals with OEMs, especially for the higher performing SKUs, which should be the ones we're most interested in.

The ASP is higher, but it's not like they get the full amount. Distributors and Retailers take a cut.

 

[krumme]

I cant see how it matters:
Tsmc is gearing up for 5nm already.
Samsung is ramping different 7nm variants.
Intel modem business was irrelevant from the get go.
Phones on the higher end starts to use less expensive chipsets.
There will be enough capacity. New nodes is expensive as hell.

 

[dark zero]

Intel has a very bad time now. When EUV hits, the one who will get screwed up hard will be Intel since the node advantage they had will be nullified.

They are on the position of AMD when they were on 28nm, but this time at 14/10nm.

 

[NostaSeronx]

My deep sources tell me than Intel will be having a 7nm 193i in risk production in the 2H of this year.

- New FinFET structure, derived from inserted-oxide FinFET (iFinFET).
- New patterning technique, derived from Tilted Ion Implantation.

7nm 193i => 7nm Low Power, for 2020. Optimized for small dies.
7nm EUV => 7nm High Performance, for 2021. Optimized for large dies. Anything built in LP7 can be co-integrated into HP7 with minimal changes. Which implies TII is used in both 193i(multi-patterned version?) and EUV(single patterned version?).

 

[IntelUser2000]

Wow. No 5G modems at all? So how many products does that mean they're going to kill before launch?

Official press release says modems for Smartphones are cancelled and the rest are under review. Since many APs use integrated LTE modems that points the reason squarely at Apple.

You'll probably see the 8160's on some ultrabooks and detachables but whether there's a future beyond that is dependent on whether Intel sees it worth continuing.

I think even for PCs, LTE modems should be integrated. If not for Lakefield, the one after should have it integrated. They have an on-package solution with HP devices already, why not have it part of the Foveros stack?