GlobalFoundries Stops All 7nm Development: Opts To Focus on Specialized Processes (anandtech.com) 114
GlobalFoundries has made a major strategy shift announcement. The contract maker of semiconductors says it is ceasing development of bleeding edge manufacturing technologies and stop all work on its 7LP (7 nm) fabrication processes, which will not be used for any client. From a report: Instead, the company will focus on specialized process technologies for clients in emerging high-growth markets. These technologies will initially be based on the company's 14LPP/12LP platform and will include RF, embedded memory, and low power features. Because of the strategy shift, GF will cut 5% of its staff as well as renegotiate its WSA and IP-related deals with AMD and IBM.
GlobalFoundries was on track to tape out its clients' first chips made using its 7 nm process technology in the fourth quarter of this year, but "a few weeks ago" the company decided to take a drastic strategical turn, says Gary Patton. The CTO stressed that the decision was made not based on technical issues that the company faced, but on a careful consideration of business opportunities the company had with its 7LP platform as well as financial concerns. On the heels of this announcement, AMD said today that it will move all of its 7nm production on both CPUs and GPUs to TSMC.
GlobalFoundries was on track to tape out its clients' first chips made using its 7 nm process technology in the fourth quarter of this year, but "a few weeks ago" the company decided to take a drastic strategical turn, says Gary Patton. The CTO stressed that the decision was made not based on technical issues that the company faced, but on a careful consideration of business opportunities the company had with its 7LP platform as well as financial concerns. On the heels of this announcement, AMD said today that it will move all of its 7nm production on both CPUs and GPUs to TSMC.
Weasel words! (Score:5, Insightful)
'Not based on technical issues' but based on...lots of bullshit...as well as financial concerns.
'Financial concerns' like a 10% yield, or some other technical disaster, so full of shit.
Re: (Score:2)
Maybe Intel and GlobalF. are forming a Cartel of Incompetence.
Re: (Score:2)
It's actually just about economies of scale. The mobile processor ecosystem simply outgrew PC processors. AMD was the first to recognize this, Intel may still be in denial.
Re: (Score:3)
Re: (Score:3)
Plus, they are all hitting the wall.
Actually, no. There are credible roadmaps already to 3nm, and 1nm is a thing (actual trace pitch several times that). Smaller than that is in research, e.g., check out nanoimprint lithography. Carbon technology is also a thing. Even without these exotics, there is still plenty of Moore's law still to come in standard lithography.
Re: (Score:3)
Love to see the sources for that, because it'd be amazing if true. Best I can find on nanoimprint lithography from the wiki is 10nm -- and that's with overlays. Toshiba got 22 nm and smaller, but no specs on how small.
Intel is having trouble with 7 nm because it's using 4 masks to get there. So, it's really using older tech with many steps to etch smaller w/ these overlays. If you have to run the silicon through the light 4 times in different positions using different patterns, you can get horrible yi
Re:Weasel words! (Score:4, Informative)
Intel is having trouble with 7 nm because it's using 4 masks to get there. So, it's really using older tech with many steps to etch smaller w/ these overlays. If you have to run the silicon through the light 4 times in different positions using different patterns, you can get horrible yields as the slightest deviation will either ruin chips or severely impact their performance.
And yet, that is exactly what TSMC is doing (i.e., no EUV) and they are supposedly ramping up volume production, though I would not bet my life on the truth of this. Intel just pushed it ever so slightly too far with about 10% smaller half pitch than TSMC, and it seems, it just didn't work out. BTW, Intel is not having trouble with 7nm because there is no such thing as Intel 7nm.
I haven't seen anything credible beyond 5 nm -- and that wasn't even using silicon as the substrate.
Google "gate all around".
Re: (Score:2)
Calling bullshit. Around 3nm is a hard limit with current technology. At that point, quantum tunneling comes into full effect, and transistors cannot function.
Re: (Score:2)
Still 3: Intel, Samsung and TSMC.
Diminishing returns never sleeps. Each generation of fabs was costing about twice the last ones. Can't go on forever. CPU/GPU is INXS for 90-99% of applications.
Re: Weasel words! (Score:1)
However, they can do it. They aren't some fabless scofflaw.
Re: (Score:1)
However, they can do it. They aren't some fabless scofflaw.
It isn't that they can't do 10nm, it's that they can't do it without losing money on everything produced. Not in a state where you would want to ramp up production.
I'm betting that GF was seeing a similar situation with 7nm, especially since it isn't exactly cheap to produce at 7nm anyway, and then your biggest customer (AMD) has a second source that is slightly ahead of you.
That means they'd be making 7nm parts in such low quantities that they wouldn't make any money at it. So like they said, they'd be bet
Re: (Score:2)
That's because:
1. Each generation represents a large relative leap to the last one (7 nm is literally 50% shrink of a 14nm, and 42% shrink of 12nm).
2. Theoretical limit of approximately 3nm is getting closer (quantum tunneling).
Add these two to all the traditional problems you get when you shrink the transistors, like issues with power and leakage, and you have an industry which is very close to hitting a wall of theorethical physics in this particular implementation.
Re: (Score:3)
Before you consider the rest of the factors, 10% yield sounds like a working process to me!
You'd know if it didn't work when the yield was well below 1%.
You won't know if 10% is good or bad unless you do a financial analysis. Lots of high tech processes have low yield.
It might not be weasel words, it might just some edujamakatid words.
Re: (Score:2)
Sure, projected yields and financial analysis come before 'disaster'...the point is they didn't just change course by themselves. Something changed, likely something technical. They learned something or a client walked away after learning something.
Re: (Score:3)
They learned that 7nm is really hard. You have your choice of way too many deep UV multipatterning steps or wacky new EUV technology still not ready for prime time. To get an idea just how not ready, google "EUV pellicles".
Re: (Score:2)
Everybody is. Building fab lines around it is another thing, they wisely backed out for the moment.
it all comes down to money (Score:2)
the fab side of AMD was always looking for another buck. nothing has changed.
Re: (Score:2)
GF bleeds money. They have always bled money. Investors want to earn money, not lose it.
The problem is how they expect to expand in the future. Slow improvements to FD-SOI manufacturing? Still need development costs, still need skilled researchers and technicians. They can't just kick them all out now and expecting them to come crawling back when GF decides they need them.
Maybe this is the only real alternative if GF is to survive in some form. Well, time will tell.
Diminished returns (Score:2)
Actually, at these scales of under 100nm, semiconductors have already hit the point of diminishing returns. After all, what are the reasons to do a die shrink, rather than just run a chip production off a fab that's already been up and running, and probably written down? It's cost, and at today's level, things like power consumption and speed are distant followers. But when it costs $10B to build a fab that can do one of these, then going from, say, a 30nm to a 14nm would not give a manufacturer the cost
Re: (Score:1)
You didn't make it clear how this relates to the story and what you want GlobalFoundries to do. Do you really want them to keep throwing good money after bad when they can't afford to fund the new 7nm node all the way to production? Is the world really better off having them do 7nm when a few other companies expect to be able to do it better?
Re: (Score:3)
So who, exactly, gets to decide what 'real happiness' is? Who gets to decide what 'improves' your life?
Your ideas aren't stopped by some mystical 'them', they are stopped because they are nothing more than 'the whole world should do what I want'.
Re: (Score:2)
Did Moore's law just end? (Score:5, Insightful)
Did Moore's law just end? Intel said they thought it had...maybe this is confirmation.
Re: (Score:2)
Did Moore's law just end?
Intel's law ended. For the big asian fabs, density continues to track Moore's law pretty well.
Re: (Score:2, Informative)
For the big asian fabs, imaginary density continues to track Moore's law pretty well.
Just because they say it's 7nm doesn't mean it is 7nm
Re: (Score:3)
Just because they say it's 7nm doesn't mean it is 7nm
You heard somebody say that, you're not exactly sure what it means but it sounds good, so you reposted it to the internet on the assumption that nobody besides you has ever heard this on the internet.</rant> "Actual" node dimensions, whatever those are, also follow a curve, maybe not exactly the same one as the nominal node name. For example, 7nm half pitch is about 40nm for the 7nm node. (Intel went for 36nm half pitch at its roughly equivalent 10nm node, just a bit too fine, with disastrous results
Re:Did Moore's law just end? (Score:5, Informative)
Firstly, their definition of "7nm" is actually, about the same as Intel 10nm, it's stupid marketing lies and speak.
Secondly, Intel themselves are stuggling like total crazy to achieve 10nm reliably.
So, to answer your question, yeah, I think Moores Law is very very close to dead if not dead. Just go look up benchmarks for processors designed 5 years ago, they're still viable now.
If you compare frequency, IPC, core count, relatively, you'll see the amount of progress we've had in the past 5 years is, atrociously bad, very, very bad.
This is why mom / pop PCs built even up to 7 years ago, just need 2 more sticks of ram, the dust blown out and an SSD with a Windows re-install, they'll be fine for another 5.
It's over, no more bleeding edge, insane fast PCs. Just very very small burps forward.
Re: (Score:2)
How is transistor density increasing when they are having extreme difficulty shrinking the processes?
The whole point of this discussion...
Re: (Score:1)
One, Moore's law is about transistor counter in a densely packed IC. To remain profitability, this implies increased transistor density but it would be sufficient to at times grow the die size to accomodate the transistor growth. Two, consumer processors today do not have 32 cores. High end server AMD processors have 32 cores. Three, if we go by transisto
Re: (Score:2)
Moore's Law in the phrasing "transistor count per unit area doubles every X months" is probably dying soon. I don't think we've reached the absolute end but we're well out of the exponential section of the curve. We've been lagging since 22nm or so, and transistor shrinks are only going to get slower and slower.
A tweaked phrasing may still be viable, though: "transistor count per unit money doubles every X months". Near the end of the long 32nm/28nm era, the GPU vendors were making some pretty massive chips
Re:Did Moore's law just end? (Score:4, Informative)
Only if you only look at Intel processors. Current ARM processors blow away the ones from five years ago. Same with GPUs from AMD and NVIDIA. AMD's current x86 processors are also way faster than their ones from five years ago. It's just Intel that's falling behind.
Re: (Score:2)
> If you compare frequency, IPC, core count, relatively, you'll see the amount of progress we've had in the past 5 years is, atrociously bad, very, very bad.
In you only consider Intel, I agree.
Luckily, AMD with Ryzen begs to differ.
I bought in with an i4670k oc'ed at 4.2Ghz. That's a 5 y old CPU with 4 cores and 4 hw threads that cost me almost £300 from memory.
Today I could buy an AMD 2700x for £300 which clocks higher and comes with 8 cores and 16 hw threads.
So assuming that software can ma
Re: (Score:2)
Compare the speed of ONE of your cores at the SAME frequency as the 5 year old CPU. You'll not only find little has changed there.
Now compare the qty of transistors, per square mm, and little has changed there either.
Re: (Score:2)
AMD not being competitive at the high end of the CPU market is what allowed Intel to sit back and do NOTHING to really advance the PC industry. The moment AMD came up with something that could really compete at the high end, Intel was forced to accept that dual-core is for low end budget machines rather than mainstream, and quad-core should have been seen as the entry level even four years ago.
Intel had processors that could run at 5GHz four years ago, and they intentionally crippled the ability to run at
Re: (Score:2)
To be fair, their focus is on what sells. What sells are laptops and servers. What they both need is low power. What the later needs is multiple cores and scalability.
So what we get produced in volume are low power multiple core processors. There isn't demand for bleeding edge speed, so they aren't building for it.
It's been trending that way for 5 to 10 years now. I started noticing when they started to have improvement margins between generations of single digits, while getting +25% power savings, as that
Re: (Score:2)
Did Moore's law just end? Intel said they thought it had...maybe this is confirmation.
Not quite yet, we have designs [phys.org] for a 4nm transistor made with silicon so I think there's room for one more iteration past 7nm. But yes the end is coming and soon, first half of the 2020s I think you can hold the funeral.
Did Moore's law just end? - Economics (Score:2)
Another FAB bites the dust (Score:2)
Re: (Score:2)
Just to clarify, this is also triple/quad multipatterning (TSMC) vs partial EUV (Samsung). TSMC will likely transition to EUV later in the 7nm node. All eyes should be on trailblazer Samsung to see if EUV is actually economical.
Very strategical. (Score:2)
Wow. Much drastical. Very strategical.
Re: Very strategical. (Score:2)
I think they meant strategory.
What? Why Now? (Score:2)
Apparently GloFo has been bleeding money for years, and they're not big enough to compete with Samsung and TSMC on the latest nodes and remain profitable. So instead of chasing the latest process shrink, they're targeting niches that are more profitable and less served by the other companies. Despite what the summary implies, they were still pretty far away from volume production of 7nm, for which they're using standard lithography tech at this point. EUV 7nm would come later and require $billions more to g
Smart (Score:5, Informative)
Smart move. There is just too much bleeding edge science and engineering at 7nm, this is a physical reality. Stick with profitable, mature fab tech and iteratively improve it. Get into 7nm when some of the horrible EUV issues have well known solutions, which should carry on to 5nm.
Meanwhile, the big Asian fabs are said to be ramping 7nm production, but as far as I know, nobody has seen actual parts arrive beyond samples. Certainly not enough to have a good idea about yields. Definitely a believe it when you see it situation. Of course, I hope that Samsung and TSMC have actually overtaken Intel at this transition, and given the economics of the situation it seems inevitable, but we do not have proof it has actually happened yet.
Re: (Score:3)
Yea, until you see that in another four years, everyone will be looking at 7nm and better fab processes, and they are left with nothing that people actually want. Zilog used to be huge in the CPU business back in the 1980s, and while still alive, it is insignificant these days. If 7nm were as close to being ready as it seemed to be at GF, dropping it right now just seems foolish. It would be like EA dropping a title that they had put millions of dollars into developing, everything was set and actually
Re: (Score:2)
There's an old addage, don't throw good money after bad.
Sometimes the smart move is to cut your losses and move on, even if you are 90% of the way there (especially if that last 10% cost more than the first 90%)
Re: (Score:2)
If you have an idea how to make a production line better than current silicon technologies in only 5 years, you'll be a billionaire.
But you don't. You're just pulling numbers out of your ass.
Re: (Score:2)
If you have an idea how to make a production line better than current silicon technologies in only 5 years, you'll be a billionaire.
But you don't. You're just pulling numbers out of your ass.
He's making room for his foundry.
Re: (Score:2)
7nm is hard (Score:3, Interesting)
7nm is hard, heck, 14, 10nm are hard, even 28nm is hard.
Many more effects, OCV, double, tripple, or quad patterning, not to mention new STA models, fault models, transistor models, extraction models, DRC, ERC and LVS models, all of these cost money.
phones will likely stay on the 28nm process node for a long long time, and unless you plan to charge $600 for a cpu, it's unlikely that even intel or AMD will go to that node for the consumer level stuff.
the ROI just isn't there.
Re: (Score:2)
The most important thing in my pants is only 2.5 nanometers in diameter.
Re: Uh? (Score:1)
Re:Intel to blame? (Score:5, Insightful)
AMD's contract with GlobalFoundries was ending this year anyway, so AMD doesn't lose anything with this announcement, other than a potential alternate source. In fact, it's entirely possible that GF's failure to secure AMD as a client may have played into their decision to drop 7nm.
That said, with only Samsung and TSMC on the leading edge now, it does mean that AMD has one less bargaining chip next time negotiations come around (i.e. they can't realistically threaten to go back to GF), whereas Intel will continue using their own processes as they always have. So, at least in that minor regard, I suppose this does benefit Intel and harm AMD somewhat.
Re: (Score:1)
with only Samsung and TSMC on the leading edge now, it does mean that AMD has one less bargaining chip next time negotiations come around
Weirdest thing would be, AMD contracts with Intel to fab some upcoming GPU. It could happen. Another thing not out of the question: Intel follows AMD and spins off its fabs. Very not out of the question.
Re: (Score:2)
Possible, but I doubt it. Intel may not be able to figure out how to get 7 nm to work for years. TSMC is the obvious choice for AMD to remain in the lead for CPUs and their only hope for parity on GPUs. TSMC may make them pay more for the best quality silicon and fastest capacity, but it'd be worth it.
Re: (Score:2)
with only Samsung and TSMC on the leading edge now, it does mean that AMD has one less bargaining chip next time negotiations come around
Weirdest thing would be, AMD contracts with Intel to fab some upcoming GPU. It could happen. Another thing not out of the question: Intel follows AMD and spins off its fabs. Very not out of the question.
Wow, now who got triggered by that? Intel crying uncle on in-house fabs, no so bad, worse could happen.
IBM loses, too (Score:4, Interesting)
IBM's fab in Vermont was sold to GF, and believe it continued to be defense-rated for (nobody's talking) type chips. so folks doing things they shouldn't in places they are not supposed to be are going to be scampering for product nobody should know about. look for Intel to suddenly get its 7nm act together.
Re: (Score:2)
That would only make sense if Essex Jct. were producing on the latest fab and they're not. They do a lot of high-volume chipsets at larger sizes - RF chipsets and game systems chips your probably have heard of or own. The ITAR stuff is minor in terms of volume.
Re: (Score:2)
it does mean that AMD has one less bargaining chip next time negotiations come around (i.e. they can't realistically threaten to go back to GF)
It doesn't say that they're never going to compete at 7nm, they're just saying that they're not going to compete right now while it's the cutting edge. From the story:
Gary Patton admits that GlobalFoundries never planned to be a leading producer of 7-nm chips in terms of volume. Furthermore, the company has been seeing increasing adoption of its 14LPP/12LP technologies by designers of various emerging devices, keeping Fab 8 busy and leaving fewer step-and-scan systems for 7LP products.
So the main business reasons seem to be related to the fact that they're getting increased demand for their older processes, and can make more money on that then on doing the R&D to be a bit player at 7nm. They'd need a new factory to do both, and they don't think it is worth building a new factory for 7nm right now.
So I would expect them
Re: (Score:3)
Lattice Semiconductor Appoints Jim Anderson (former AMD General Manager and Senior Vice President of the Computing and Graphics Business Group. ) as CEO
https://www.marketwatch.com/pr... [marketwatch.com]
Lattice Semiconductor Corporation LSCC, a leading provider of customizable smart connectivity solutions, announced the appointment of Jim Anderson as the Company’s President and Chief Executive Officer, and to the Company’s Board of Directors, effective September 4, 2018. Mr. Anderson brings broad technology indus
Re: (Score:2)
T
Re: (Score:2)
Doubt it, since AMD just shifted to TSMC.
Probably more like GlobalFoundries wasn't confident in the progress they were making, so cut their losses.