Discussion What if Intel skipped their internal shrink node factories and went with TSMC?

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

CentroX

Senior member
Apr 3, 2016
351
152
116
Is this possible? Just to catch up with AMD. How long can their internal process struggle before they take action?
 

pedreiro77

Junior Member
Dec 13, 2019
1
0
6
I will be as straightforward as I can and I don't want to get into pointless discussions with people who don't pay attention to the facts. I will round up most numbers because it will consume a lot of time to be extremely precise, and 2019 attention spans are shorter.

TSMC Spent about 15 billion USD in order to move from 12nm to 7nm [Source] [Source2].

Intel was extremely comfortable in 2015 (obvious).

Intel PROFIT (not REVENUE) is about 12-16bi/year (just look up the data from their own reports).

Intel does have some keen interest in everything that is from Israel, not a conspiracy, it is simply a fact. Moore is jewish, they invest a lot of money in Israel, their previous CEO that was quite dumb not to say a scammer a.k.a. Brian Krzanich is also jewish. When Intel needed to make some fake benchmarks very recently in order to sabotage AMD's image, they hired some guys from Israel to do the fake benchmarks.

In 2015 Brian Krzanich blew 15 BILLION dollars (a lot of people in the board were agaisnt this decision) to buy an Israeli company called MobilEye (yes, the same 15bi they desperatedly need right now). In their own words, they made this huge decision of blowing huge ammounts of cash buying a company that works with "SELF DRIVING CARS" technology, was to futureproof them because this thing > self driving cars is going to be big in 2030 (yeah, let's think 15 years into the future and forget about immediate stuff because we can feed people 4C/8T CPU's forever and they will buy it, because AMD sucks and AMD processors are hot and use lots of energy).

Turns out that buying this company was a TERRIBLE IDEA. Their self-driving tech sucks and is being outclassed by open source stuff. The company was OVER EVALUATED by a factor of 20+. The company was worth less than 700million, they paid 15bi for that.

Rumor has it that Brian had profited from this shady, dumb deal.
Everything is out there, its just up for us to see, but yeah, anyway, the 9900K is the best for gaming right? so who needs AMD.
 
Last edited:

A///

Diamond Member
Feb 24, 2017
4,352
3,154
136
All the huge factories that produce high tech goods for western companies on mainland China are run by either Americans or Europeans for a reason. The Chinese just don't have the knowledge to produce these things on their own. They can make cheap, inferior knock-offs but that is it.
I've heard this and I'm willing to bet it's true. It was asinine for the other user to suggest TSMC a Taiwanese company is feeding China when China can't get themselves in order and must rely on badly botched copies of western goods or other Asian goods.

My only experience with fabs was working in them for two companies I'd prefer not to disclose. Security at both were very, very tight. Security at TSMC and even Intel, I'd imagine, is even tighter. Probably wouldn't be able to let out gas without them knowing it was you.

And before anyone makes that joke, no, a giant ball of puffy gas doesn't develop in your clean suit/bunny suit and follow you around all day long until you remove it. This may have or may not have been tested many years ago...


Inferior knockoffs are possible. I seem to recall knockoff Intel HEDT parts, albeit from older generations, coming alive on certain Chinese goods sites. I've read and been told many theories on this including Intel themselves handing over the equipment but none make reliable sense. It's way easier to buy old lots of processors and reetch the writing on the processor.
 
Last edited:

Adonisds

Member
Oct 27, 2019
98
33
51
I don’t know how you can be so sure of that. 7nm doesn’t use SAQP, which is one of the biggest problems with 10nm DUV. I believe by now, that Intel has learned what it needed to from other aspects of the 10nm node. Node implementation is incrementally dependent with multiple teams in different stages of research and development. If spending on EUV equipment doesn’t go way up over the next 18 months - that would be very bad news.
Does Intel divulge their EUV equipment spending in their earnings?

Could Intel have moved to 7nm sooner or was the EUV equipment necessary not available in good quantity?
 

JasonLD

Senior member
Aug 22, 2017
485
445
136
Does Intel divulge their EUV equipment spending in their earnings?

Could Intel have moved to 7nm sooner or was the EUV equipment necessary not available in good quantity?


Intel has started placing equipment and materials orders for EUV fabrication processes since August, and is stepping up its pace of orders, according to industry sources.
 
  • Like
Reactions: Ajay and Adonisds

ksec

Senior member
Mar 5, 2010
420
117
116
Does Intel divulge their EUV equipment spending in their earnings?

Could Intel have moved to 7nm sooner or was the EUV equipment necessary not available in good quantity?

The single bottleneck in the industry. I wrote this on SemiWiki in 2017/2018, why would anyone expect Intel to be on time with their 7nm when everyone is relying on ASML for their TwinScan Machine. It doesn't make much sense.

At the time ASML analyst were stating ( I think the word promising would be better ) that ASML would ramp up their production in time. Cough.

Both Samsung and TSMC are trying to get as much of them as they can. TSMC have higher demands than expected, ( they are conservative with their estimate as usual, so not a surprise ) Thanks to AWS and AMD, possibly more clients we dont know yet. Samsung is trying to get their capacity up and running for Qualcomm as well as using EUV for DRAM and NAND. Micron are thinking of EUV as well,

I was surprised where will Intel get enough EUV for their 7nm in time for 2022? May be another 10nm ramp like they are currently doing in limited quantities? ( Forget about their 2021 target, they are aiming for Q4. Which is more like 2022 in volume )

ASML initially targeted 30 Unit for year 2019. Unofficially they were actually targeting 36, but 30 was a safe bet. And as it turns out, ( If I remember correctly ), they only shipped 26. Some would argue the C revision improves capacity, but in reality for like of Samsung and TSMC those factors were known before hand and were factored into their capacity calculations already.

So what we have now, is an increase in demand of EUV wafers, and therefore increase demand of EUV Machines, with lower than expected supply for them.
 
Last edited:

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
IIRC, Intel pumped something like a billion dollars into ASML to help move EUV development along. I wonder I’d this might affect their allocation %?

Actually, it was $4B US. TSMC and Samsung put in 100s of millions as well.
 
  • Like
Reactions: Adonisds

moinmoin

Diamond Member
Jun 1, 2017
4,934
7,619
136
All the huge factories that produce high tech goods for western companies on mainland China are run by either Americans or Europeans for a reason. The Chinese just don't have the knowledge to produce these things on their own. They can make cheap, inferior knock-offs but that is it.
Good point, let's talk Electronics Manufacturing Services, those are the ones doing the actual production in many cases. You're right those are most often not Chinese. What are they?
Wikipedia list following companies in this sector as having an annual revenue of over US$3 billion:
I don't think you can just jump to the next node like previous node never existed. Pretty sure there are parts of previous node design that will carry over to the next one. if 10nm gets delayed, all future nodes will inevitably gets delayed.
To a point, but one should always invest in concurrent R&D independent from each other so they aren't blocking each other. Keep in mind Intel had a biannual node cadence and want to have that again. 22nm was in 2011, 14nm was in 2014 (wasting time of half a new node), 10nm is counted as 2019 (wasting time of two and a half new nodes), 7nm is supposedly coming at the end of 2021, but more like 2022, and even that seems optimistic to me (so maybe wasting time of half a new node or more). That, by Intel's own preferred cadence, add up to more than 3 new nodes that they missed by having their work on the current three nodes 14, 10 and 7nm block their traditional progress.

Regardless, Intel wouldn't be in a bad position if they manages to move their consumer/Server CPU lineup to 7nm in 2022. So all the talk about Intel spinning off their fabs or fabbing their CPUs on TSMC is bit premature at this point.
7nm in 2022 is not a bad position? Of course if it's competitive it's better than the state right now. But from Intel's POV, their latest process node merely being competitive is what I would call bad. We'll have to see how that all plays out, but my trust in Intel's capability to execute new nodes is pretty much in shatters right now.

I don’t know how you can be so sure of that. 7nm doesn’t use SAQP, which is one of the biggest problems with 10nm DUV. I believe by now, that Intel has learned what it needed to from other aspects of the 10nm node. Node implementation is incrementally dependent with multiple teams in different stages of research and development. If spending on EUV equipment doesn’t go way up over the next 18 months - that would be very bad news.
Intel had 1 year warning with 14nm's delay as well as 5 years of nonsense time wasting with still no server and desktop chip in sight with 10nm. If their next node working well depends on their spending on EUV equipment over the next 18 months they have already lost.
 

JasonLD

Senior member
Aug 22, 2017
485
445
136
7nm in 2022 is not a bad position? Of course if it's competitive it's better than the state right now. But from Intel's POV, their latest process node merely being competitive is what I would call bad. We'll have to see how that all plays out, but my trust in Intel's capability to execute new nodes is pretty much in shatters right now.

Considering Zen 4 is late 2021/2022 product, I wouldn't say 7nm in 2022 is a bad position, considering 7nm is competing node against TSMC's 5nm. Intel is currently behind on process but not so much in terms of architectures, and their future plans is to make sure their architecture is no longer tied on process node, so they actually have a plan B this time. So we won't see the repeat of endless skylake architecture on 14nm again.

Intel's current plan is to have majority of their portfolios on 7nm ready by 2022 so it must be their priority alongside with fixing their 10nm. While It is right to be skeptical about Intel's future plan with all the troubles they had with 10nm. most of 10nm troubles they had doesn't look like it will apply with 7nm with EUV. So I am still on 50/50 in terms of Intel's ability to deliver 7nm on 2022 on majority of their CPUs.
 

Greyguy1948

Member
Nov 29, 2008
156
16
91
Today AMD is small compared to Intel. EPYC has less than 5% of the server market. But if they grow up will TSMC have capacity enough or will they have to go back to 12 nm at GF?
 
Last edited:

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
To a point, but one should always invest in concurrent R&D independent from each other so they aren't blocking each other. Keep in mind Intel had a biannual node cadence and want to have that again. 22nm was in 2011, 14nm was in 2014 (wasting time of half a new node), 10nm is counted as 2019 (wasting time of two and a half new nodes), 7nm is supposedly coming at the end of 2021, but more like 2022, and even that seems optimistic to me (so maybe wasting time of half a new node or more). That, by Intel's own preferred cadence, add up to more than 3 new nodes that they missed by having their work on the current three nodes 14, 10 and 7nm block their traditional progress.

7nm in 2022 is not a bad position? Of course if it's competitive it's better than the state right now. But from Intel's POV, their latest process node merely being competitive is what I would call bad. We'll have to see how that all plays out, but my trust in Intel's capability to execute new nodes is pretty much in shatters right now.

Intel had 1 year warning with 14nm's delay as well as 5 years of nonsense time wasting with still no server and desktop chip in sight with 10nm. If their next node working well depends on their spending on EUV equipment over the next 18 months they have already lost.

1. Intel does work on more than one node at a time, however, they are dependent on milestones being reached in the node before them. If a critical milestone is missed in node x, it can delay the development of node x+1. This is called node learning. Obviously, had problems with 14nm, and doubled down on 10nm being overly aggressive. Intel has had extra time to develop 7nm - we'll see if this bears fruit.

2. Yeah, 7nm in 2022 will likely match TSMC 5nm; that’s a fair match up. TSMC will be in HVM on their 3nm in 2023, that's a problem for Intel because AMD will have that advantage.

3. How many EUV lithography machines will Intel need? They aren’t using EUV on all layers, the rest will be on existing DUV equipment. @ksec made a great post on ASML Twinscan equipment production - this is a problem for everyone. These equipment purchases were signed well in advance of production, so we'll have to wait and see what Intel’s allotment will be. Maybe they bought up 70% of ASML's 2020/2021 production - we have no way to know right now. (I should have said over the next 24 months instead of the next 18). Anyway, too many variables to know how things are going right now. TSMC is confident that their more heavily EUV dependent 5nm node will be on time - under the same logic, we should be questioning their ability to deliver.
 

ksec

Senior member
Mar 5, 2010
420
117
116
IIRC, Intel pumped something like a billion dollars into ASML to help move EUV development along. I wonder I’d this might affect their allocation %?

Actually, it was $4B US. TSMC and Samsung put in 100s of millions as well.

Damn I knew this was going to come up, but I couldn't be bothered to included this in my previous rant-ish reply.

The link was from 2012, here is one from 2018


If I remember correctly none of the current Foundry has stake in ASML any longer.

I do want to point out those Intel order were made in advance, whether ASML will ship them on time, and whether Intel can get 7nm yield right ( TSMC had 7nm as a test case along with their own toolchain ) is entirely different matter.
 
Last edited:

CHADBOGA

Platinum Member
Mar 31, 2009
2,135
832
136
Zero evidence but I still feel Intel has some shenanigans. Something like a core2duo change coming in 2020
Intel's problem this time round is not the design of their CPU, but their manufacturing process, so unless that gets fixed, it is unimaginable that they could have another Conroe to bail them out.
 
  • Like
Reactions: moinmoin

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
Damn I knew this was going to come up, but I couldn't be bothered to included this in my previous rant-ish reply.

The link was from 2012, here is one from 2018


If I remember correctly none of the current Foundry has stake in ASML any longer.

I do want to point out those Intel order were made in advance, whether ASML will ship them on time, and whether Intel can get 7nm yield right ( TSMC had 7nm as a test case along with their own toolchain ) is entirely different matter.
Thanks for the update. I wasn’t linking AMSL's deliveries and the success of Intel’s 7nm EUV process - it just affects whether or not they can produce in volume.
 

Roland00Address

Platinum Member
Dec 17, 2008
2,196
260
126

From the Article: Intel chips will be fabricated on TSMC's 7nm optimised version of its 6nm process. (I'm not sure if that means TSMC N7P, N7+, or N6.)
Remind me more technical people than me. N7P is just 7nm but a little more power efficient or a little faster and we are talking 10% best case on those two metrics.

N7+, and N6 are similar on the density front but it is other characteristics where they diverge? What are those other characteristics? Both N7+ and N6 are 15 to 20% more dense than N7P.

Is my recollection correct?
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
@Roland00Address

I don't think N6 is an EUV node. It has the same design rules as N7/N7P and is essentially a "dead end" for customers that don't want to pay for N7+ or N5
I thought a chip maker could go from N7->N6->N5(EUV) as a lower cost option (incremental changes to the design rules).
My Bad, you were right:
wikichip_tsmc_logic_node_q1_2020.png
 
  • Like
Reactions: Tlh97 and Saylick