Tuesday, September 4th 2018

Analyst Firm Susquehanna: "Intel Lost its Manufacturing Leadership"

Intel was once the shining star in the semiconductor manufacturing industry, with a perfectly integrated, vertical product design and manufacturing scheme. Intel was one of the few companies in the world to be able to both develop its architectures and gear their manufacturing facilities to their design characteristics, ensuring a perfect marriage of design and manufacturing. However, not all is rosy on that field, as we've seen; AMD itself also was a fully integrated company, but decided to spin-off its manufacturing arm so as to survive - thus creating GLOBALFOUNDRIES.But Intel was seen as many as the leader in semiconductor manufacturing, always at the cutting edge of - well - Moore's Law, named after Intel's founding father Gordon Moore. Now, Mehdi Hosseini, an analyst with Susquehanna, has gone on to say that the blue giant has effectively lost its semiconductor leadership. And it has, in a way, even if its 10 nm (which is in development hell, so to speak) is technically more advanced than some 7 nm implementations waiting to be delivered to market by its competitors. However, there's one area where Intel will stop being able to claim leadership: manufacturing techniques involving EUV (Extreme UltraViolet).
It's being reported that intel has decided to postpone its EUV efforts to other processes, not taking it up on its 7 nm process development. This saves Intel money in adding yet another technique and technology to an already hard to manufacture node, but leaves the door ajar for the likes of TSMC and Samsung (of which the latter is expected to have the more complex EUV implementation, in more layers, at least in the beginning). TSMC is looking to develop both 7 nm and 7 nm+ manufacturing processes, where only the latter will feature EUV integration - a way to divide costs and reduce reliance on a still exotic technology. But while Samsung and TSMC are both looking towards some level of EUV integration by 2019, intel is looking towards a farther 2021.
According to Susquehanna's Mehdi Hosseini, TSMC is the company to look out for as the semiconductor manufacturer leader, as it "appears to be winning most of the leading-edge design wins due to better 7nm process technology performance, lower power consumption and better area density." Indeed "the times, they are a changin'." Source: EETimes
Add your own comment

61 Comments on Analyst Firm Susquehanna: "Intel Lost its Manufacturing Leadership"

#26
ironcerealbox
I read the actual sourced page. I recommend reading it. Raevenlord is just simplifying the readability of the actuall EETimes article (though, it is not necessarily shorter). I disagree with the style of the headline though but that is for another time.

Basically, Intel is willingly giving up on EUV and will pursue other avenues of manufacture. That leaves TSMC and Samsung. GloFo (Global Foundries) left or gave up on 7nm EUV (which they were a partner with Samsung). In my opinion, GloFo was seeing that industry leaders were not happy with their and Samsung's 7nm 8-to-10 layer EUV process and decided to ditch the 7nm process altogether.

Samsung's design is aggressive since their 7nm requires EUV and 8-to-10 layers and will, supposedly, arrive after TSMC's 7nm non-EUV. This is backfiring on them but they are still willing to push through (even at the cost of alienation). The costs are high (which have to be passed on).

TSMC will have 7nm non-EUV and 7nm+ EUV with the former arriving before Samsung's 7nm EUV process and the latter arriving after Samsung's 7nm EUV process. Industry experts are saying having flexibility with both types is an advantage, according to Mehdi Hosseini (the analyst from Susquehanna).
Posted on Reply
#27
hat
Enthusiast
Dante Uchiha said:


7nm seems superior. I wonder what will happen when both AMD (7nm) and Intel(10nm) have CPUs with clocks near the 5Ghz barrier. Are we going to start the Battle of IPC improvement or will the CPUs market stagnate? Would a "simple lithography change" bring IPC gains?
Hard to say. Both Intel and AMD will be looking for ways to increase performance. Part of that could be making the processor faster clock per clock, or raising clock speeds, or something else entirely, like that EDRAM that was on the 5775c.

IPC won't change if you only change from 14nm to 10nm, for example. It usually does though when that happens, because at least some minor changes are made to the design, and it might clock higher too. 65nm and 45nm Core 2 Duo/Quad was a good example of this.
Posted on Reply
#28
oxidized
And we should take his words as facts, because?
Posted on Reply
#29
remixedcat
the name of that firm matches the name of a city that's gonna do some nuke prep test or somethin o_O wierdness.
Posted on Reply
#30
Prima.Vera
Dante Uchiha said:


7nm seems superior. I wonder what will happen when both AMD (7nm) and Intel(10nm) have CPUs with clocks near the 5Ghz barrier. Are we going to start the Battle of IPC improvement or will the CPUs market stagnate? Would a "simple lithography change" bring IPC gains?
Aye. Saw those specs over other forums too. Both Intel's 10nm and the rest of Asian's 7nm seems to be the same on average. Have no idea why Samsung and TMSC are calling their tech 7nm, when it was proved that their density is actually higher than 10nm, even than Intel's.
I guess there is no penalty to just use numbers for marketing purposes when billions are involved. If only intel would have said that their 10nm tech it's actually 7nm, nobody would have bat an eye.
Posted on Reply
#31
Frick
Fishfaced Nincompoop
oxidized said:
And we should take his words as facts, because?
You don't, but one could choose to use it as a guideline.

And about the dude not knowing the industry he's just been doing this for more than ten years, plus he worked at NSC and he has an MS in electrical engineering. I rather take his word for it than some random internet dudes.
Posted on Reply
#32
notb
Dante Uchiha said:

7nm seems superior. I wonder what will happen when both AMD (7nm) and Intel(10nm) have CPUs with clocks near the 5Ghz barrier. Are we going to start the Battle of IPC improvement or will the CPUs market stagnate? Would a "simple lithography change" bring IPC gains?
Why would AMD suddenly have clocks near 5GHz? Zen+ hits something like 4.5GHz - already an improvement compared to Zen.

Going to a smaller node, we should expect lower clocks (at least in the first revision).

Intel, on the other hand, takes care of high clocks from the start, because that's how they differentiate the lineup. I think this is one of the reasons why 10nm is so delayed. They need it to hit near 5GHz boost from the start, or they won't be able to build a typical Intel-ish lineup.
Posted on Reply
#33
Manoa
not surprised that intel has higher density silicon, they after all have custom fabrications, this becomes eventally worth, and this density show it, but it also explain why intel didn't jump ship and abandon their custom designs and buy the silicon from TSMC or global foundaries: there is no real big difference. but I would worry about intel is 1 year later
it sad that global foundaries spended billions $ in this 7nm and buy 2 ASML buses of it and now just stop it all
can someone expert explain what is quantum problems and why it so much harder to go smaller into 7nm ?
Posted on Reply
#34
nemesis.ie
Prima.Vera said:
Aye. Saw those specs over other forums too. Both Intel's 10nm and the rest of Asian's 7nm seems to be the same on average. Have no idea why Samsung and TMSC are calling their tech 7nm, when it was proved that their density is actually higher than 10nm, even than Intel's.
Higher density = better, i.e. cramming in more transistors in the same space.
Posted on Reply
#35
seronx


Oof~

By 10nm++ FinFET from Intel, Samsung will be ramping up 3nm GAA EUV. TSMC's 3nm Fab($20B) won't be up and running at worst till 2022.
Posted on Reply
#36
R0H1T
seronx said:


Oof~

By 10nm++ FinFET from Intel, Samsung will be ramping up 3nm GAA EUV. TSMC's 3nm Fab($20B) won't be up and running at worst till 2022.
So TSMC's jumping from 7nm straight to 3nm, or is that more marketing BS than ever before?
Posted on Reply
#37
seronx
R0H1T said:
So TSMC's jumping from 7nm straight to 3nm, or is that more marketing BS than ever before?
TSMC is..
7nm FinFET 1H2018
7nm FinFET Plus 2H2018
5nm FinFET 2H2019
3nm ___ sometime 2022

They aren't jumping, but their 7nm is short-lived.
Posted on Reply
#38
R0H1T
seronx said:
TSMC is..
7nm FinFET 1H2018
7nm FinFET Plus 2H2018
5nm FinFET 2H2019
3nm ___ sometime 2022
This is the proposed roadmap, right?
Posted on Reply
#39
seronx
R0H1T said:
This is the proposed roadmap, right?
That is their current "official" roadmap.

http://www.tsmc.com/english/dedicatedFoundry/technology/5nm.htm
"It is scheduled to start risk production in the second half of 2019."

In their MPW it points to mass-start(risk) for 7nm FinFETs at TSMC.
www.zgcicc.com/mpw/2018TSMCCyberShuttleServicePlan.pdf
7nm FinFET tape-in => Mar-26 2018
7nm FinFET plus tape-in => Aug-15 2018
etc.

Risk production for Samsung is when MPWs start. The same is for GlobalFoundries.

So, Samsung at least has a two year lead on TSMC. Since, the 3nm Fab for TSMC doesn't get finished till end of 2021/early 2022.
Posted on Reply
#40
notb
Manoa said:
can someone expert explain what is quantum problems and why it so much harder to go smaller into 7nm ?
Yeah, explaining quantum mechanics in a forum comment... :D

There are different quantum effects making problems in transistors, but IMO the easiest one to understand is quantum tunneling (or at least the general idea).
I'll skip the whole probability theory, waves, potentials and so on...

Basically, you have to remember that the world around you is not corpuscular (solid, made of small particles that collide). It's made of interacting "points".
In other words: particles are points (no size), but since they repel each other, they create a structure which has dimensions (just like a triangle is defined by 3 points).

So you have your typical macro-scale wire and there's an electric current inside, like this:
-----------------------------------------
outside....inside....outside
-----------------------------------------
............|-|....e....|-|................
-----------------------------------------
It's isolated by some walls |-|, so you can hold it and so on. In the scale you're looking at them with naked eye, they seem solid.

But as you zoom to nanometres (e.g. single transistors), the walls can't be treated as solid anymore.
Particles that create the walls repel electrons inside, but - as the walls become thinner and thinner - repelling becomes weak. As a result, sometimes an electron gets through the wall. :)

In other words: if you could "measure electricity" on a nano scale, you would also get some reading outside the walls as well:
-----------------------------------------------------------------------
outside.........wall.........inside........wall........outside
-----------------------------------------------------------------------
........e'........|----------|........e........|----------|......e'.......
-----------------------------------------------------------------------

:)
Posted on Reply
#41
Manoa
thank :)
if I understand right, it meens that the walls insulator have to be small so that you can put many transistors and less insulators into the chip
but despite the insulator is not-conductive, the electrons flow through them anyway - not because they are not-conductive (which don't exist) - but because they are very few and very small
you meen to say it crates an effect like a static charge: 10,000 volts of static still attract each other through air at distance of 10 cm - air which is not conductive
so when this things are so littel - the problem become mutch worse, and that charge is not static
but then because electrons are flowing through the walls, they destroy the walls !
I understand now I think, there is no material in the world that can be non-conductive when he is so littel :x
so this meens, the more space you spend on insulator, the less space you have for everything :x

is this how they do QA on processor logics ? they test them for "short circuits" ?
what about making the other sides of the wall charged such that there is no differential potential so the electron don't wanne flow towards outside the wall ?
Posted on Reply
#42
DeathtoGnomes
notb said:
As someone already mentioned: the analyst doesn't really know much about semiconductor manufacturing. But that's a smaller problem.

The bigger problem is what's happening with AMD stock right now. AMD hasn't updated plans, they haven't surprised us with a new product or anything. The only unexpected piece of news about AMD lately is that they lost one 7nm partner.
Yet, the stock keeps rising based on recommendations. It's +11% today because another US-based financial company decided to write a text about 7nm. :-o
Just look at the evolution of price targets:
https://www.benzinga.com/stock/amd/ratings

I wonder who's buying. It seems a lot of AMD followers started investing... ;-)
Just think what would happen if TSMC also announced delay in 7nm introduction...

If only forum members spend so much time studying computer science or business as they spend studying gaming benchmarks...
IMO the non-gaming news are only generating chaos here. I wonder how many gamers on TPU bought a Ryzen because Zen is so great for workstations... :p
Dont assume stock performance is because of a text. it makes you look bad.
Posted on Reply
#43
hat
Enthusiast
notb said:
Why would AMD suddenly have clocks near 5GHz? Zen+ hits something like 4.5GHz - already an improvement compared to Zen.

Going to a smaller node, we should expect lower clocks (at least in the first revision).

Intel, on the other hand, takes care of high clocks from the start, because that's how they differentiate the lineup. I think this is one of the reasons why 10nm is so delayed. They need it to hit near 5GHz boost from the start, or they won't be able to build a typical Intel-ish lineup.
wat

There's no reason to believe that AMD can't hit 5GHz with Zen 2 on the 7nm process. We don't have any idea what it might or might not do. Though moving to a smaller process resulting in lower clocks seems completely bonkers, at least if it's a simple die shrink with maybe a few tweaks here and there (like Wolfdale was to Conroe). I could see that happening if while moving to a smaller process you also developed a radically different architecture, like the Core 2 Duo was to the Pentium, though late Pentiums and early C2D were both 65nm.
Posted on Reply
#44
SupremeCommander
notb said:
As I said earlier. Just find a comfortable seat. This will be fun, I promise. :)

"TSMC intends to introduce a more advanced 7nm fabrication process that will use EUV for critical layers, taking a page from GlobalFoundries’ book (which is set tp start 7 nm with DUV and then introduces second-gen 7 nm with EUV). "
https://www.anandtech.com/show/11337/samsung-and-tsmc-roadmaps-12-nm-8-nm-and-6-nm-added/2

This might not be the last page from GF's book that they're going to take. :)
However, the same article states that high-volume manufacturing of 7nm EUV won't be ready until H2 2019:
TSMC plans to start risk production of products using its CLN7FF+ in Q2 2018 and therefore expect high-volume manufacturing (HVM) to begin in H2 2019.
That sounds like 7nm Vega and 7nm Epyc will be manufactured without EUV.

seronx said:
TSMC is..
7nm FinFET 1H2018
7nm FinFET Plus 2H2018
5nm FinFET 2H2019
3nm ___ sometime 2022

They aren't jumping, but their 7nm is short-lived.
As stated by the AnandTech article above, those dates just mark the start of the so called "risk production" process. High-volume manufacturing of 7nm EUV won't be ready until H2 2019, while high-volume 5nm is not due until H2 2020.
Posted on Reply
#45
Totally
Gungar said:
The "analyst" says that based on the use of EUV?!?
This analyst is only stating the obvious once this info went public. That manufacturing process was the only firm lead they had on Samsung, not that they've relinquished that it is only a matter of time. I'm myself that I didn't buy their stock when I was thinking about it during the apple lawsuit, even now despite it cooling down, I still would like to but as much as I could but it's an awkward time for me to do so.
Posted on Reply
#46
trparky
Their stock has definitely taken a beating since June 1st.

To lose $10 a share... Ouch. Investors aren't happy with Intel right now, that's for sure.
Posted on Reply
#47
Prima.Vera
trparky said:
Their stock has definitely taken a beating since June 1st.

To lose $10 a share... Ouch. Investors aren't happy with Intel right now, that's for sure.
Best time to buy shares, or to wait some more?? Decisions, decisions....
Posted on Reply
#48
Spirit
lynx29 said:
and yet, Intel is still the best gaming chip there is, and will be dominating AMD in min-FPS across the board even vs the 3rd gen ryzen next year (most likely) but eh i suppose it doesn't matter anymore.
Who cares? My 2600x regularly reaches 5ghz XFR all by itself, no manual overclock needed. It is never a bottleneck in any of the dozen or so games I play and costs all of $230. Performance upgrades for the next few years will only require a bios upgrade and a new processor. Try that with intel!
Posted on Reply
#49
WAXAs
lynx29 said:
and yet, Intel is still the best gaming chip there is, and will be dominating AMD in min-FPS across the board even vs the 3rd gen ryzen next year (most likely) but eh i suppose it doesn't matter anymore.
Min FPS also it depends of video card. Actually this is difference 5-10 percent and I think with new video cards it still will be 5-10. And for example if you're streaming in good quality 8700k freezes and you cannot play. Ryzen -not.
Posted on Reply
#50
Vayra86
In other news, water is wet...

seronx said:


Oof~

By 10nm++ FinFET from Intel, Samsung will be ramping up 3nm GAA EUV. TSMC's 3nm Fab($20B) won't be up and running at worst till 2022.
Don't believe every roadmap you see. Its easy to predict something, that's why they're called predictions. But between floods, viruses and corporate spying and sabotage, you just know this will never happen. There is no economical reasoning behind it either. Flying fast forward with new nodes every year gives way too little time to recoup expenses. Every node you don't get your moneys' worth out of, is a wasted one essentially.

Even 7nm, I think we will see some bumps in the road for, still.

The bottom line with roadmaps is that you can put 'progress and leadership in X' on top of a powerpoint slide so you can snag some investors and keep your stock afloat.
Posted on Reply
Add your own comment