Question Intel Q2: 7 nm in bad shape

Page 7 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

A///

Senior member
Feb 24, 2017
201
75
71
After their earlier troubles, they changed their structure. They begin research on new nodes nearly a decade out. They began 2nm research not long ago and that isn't due until at least 2029 or 2030 provided we, humans, are still alive by then. TSMC's conservative approach is what's keeping them from fumbling like Intel who thought they were invincible. 10nm is over 5 years later, and 7nm was due in 2017.
 

beginner99

Diamond Member
Jun 2, 2009
4,539
885
126
10nm is over 5 years later, and 7nm was due in 2017.
10nm is more than just late, it's still broken. That's why there is no Icelake-server yet. Because the yields are still bad which can easily be seen in the huge drop of margins.
Issue is you can falter once. Even this 6-months 7nm delay still would keep intel in the game, if it actually is true. If 7nm is just as bad a 10nm, hell even a little less bad, intel is toast or at least their manufacturing part. 10nm was so broken, intel bet on 7nm and expanding 14nm. 10nm has rather low capacity. It's for now just dual and quad core mobile products and not even the whole market, just a fraction thereof. Intel knows the process is broken so they limited investment in 10nm fabs. I'm just repeating another poster by the way. So if 7nm fails again they can't ride it out with 10nm because that simply doesn't have the capacity and still has terrible yields barley working for dual and quad cores.
and 14 nm in 2022 or even 2024 will simply not lead to competitive products.
 

traderjay

Member
Sep 24, 2015
166
122
116
TSMC can get into trouble with a node but will weather the storm better than intel because they don't have a bloated organization like intel with huge SG&A cost. I see that people here mention Raja quite alot but don't expect him to pull a miracle. He was okay during the ATI days during the Radeon 9000 series and things started go downhill after that. He isn't well liked at AMD either after his return both in his leadership style and technical skills. People said he is more concerned about producing the next big blockbuster at Bollywood rather than focusing on product.
 

DrMrLordX

Lifer
Apr 27, 2000
16,023
4,970
136
TSMC is not immune from trouble.
Definitely not. They could stagnate as well.

More to the point, if all of TSMC's competition fell away it would remove a lot of the incentives for them to keep pushing to try meet Moore's Law (or 1.7x of Moore's Law, at least) with a new process every two years. Remember how most people were rooting for AMD to do better so Intel wouldn't act like a fat lazy monopolist who didn't need to compete? There's nothing stopping TSMC from doing the same if they got into the same situation.
There's really no 'if' anymore, unless you count Samsung; regardless, it is TSMC's customers that will continue to push them, if nothing else. TSMC finances node improvements on the back of their customers. If they demand node improvements and pay for it, they'll get it.

Sure, Apple, AMD, Qualcomm etc. might want them to keep pushing and get annoyed if TSMC slows things down, but they wouldn't have anywhere to go so they'd be forced to accept it along with all the rest of TSMC's customers.
Nah, TSMC will just make them pay to play. Costs might go up, but the pace of advancement won't slow down on that account unless the higher costs associated with node improvements scare away customers. Also, TSMC arguably hasn't had serious competition since their 16nm node, so what you're decribing has more-or-less existed for some time.

What's changing now is that there may not be any more competition for TSMC's customers, which is the big that will change.
 
  • Like
Reactions: Etain05 and Tlh97

Arkaign

Lifer
Oct 27, 2006
20,516
971
126
I feel almost like this is the only way for Intel to get fixed. Literally sink or swim. With all the failure to execute and underwhelming product for years now, they were able to enjoy high profit and stock value almost continually, which I think deeply contributed to their lackadaisical attitude and weakness.

Well they're not smiling anymore, that's for sure. It will be interesting to see what steps they take now that the merde has really hit the fan.
 

Doug S

Senior member
Feb 8, 2020
250
301
96
After their earlier troubles, they changed their structure. They begin research on new nodes nearly a decade out. They began 2nm research not long ago and that isn't due until at least 2029 or 2030 provided we, humans, are still alive by then. TSMC's conservative approach is what's keeping them from fumbling like Intel who thought they were invincible. 10nm is over 5 years later, and 7nm was due in 2017.
Who says 2nm isn't due until 2029? That's the next step after 3nm, which is scheduled for mass production in H2 2022 (i.e. exactly the timeline you'd expect for Apple to use it for the A16) If it takes TSMC 7 years to take the next process step after N3, Intel will have ample opportunity to catch up and retake the lead!

Not saying I would be shocked if they can't maintain the two year cadence and N2 and beyond might take a bit longer, but there's a big difference between two years and seven. Typically TSMC talks about the next process when they have the current one put to bed - they knew they had N5 licked this spring, thus they started dribbling out information about N3. In 18-24 months we should expect to begin hearing things about N2, if not well...

The basic research that goes into this is really split into two tracks. One is taken by the toolmakers, or really just ASML anymore since it looks like there's only room for one EUV tool manufacturer. Their goal is to produce equipment that allows fabs to draw smaller structures at sufficient yield and sufficient throughput. It looks like the way forward for the rest of the decade is EUV using progressively higher NAs, which may eventually require exotic/metamaterial lenses, which may fall to some of their partners/vendors like Canon. That's one end of it.

On the other end you have fabs that do research to determine "given the ability to draw smaller structures, how can we use that to draw more transistors in the same area, using less power and/or with higher performance?" When current techniques run out of gas, like planar transistors did and FinFETs are on the verge of doing, they have to develop new transistor structures. Also new materials like when they switched to copper wiring, the ongoing research into graphene, that sort of thing. That's the other end of it.

When TSMC researches "2nm" - i.e. it has been named - they have already figured out what it will be and how they want to do it, or at least narrowed it down to a few candidate approaches (which all work in the lab but they'll choose from based on how well they work in mass production) This isn't pathfinding research, it is manufacturing research - how do we mass produce something that meets the criteria of what we've defined as "2nm" at sufficient yield/cost.

Pathfinding research is what you do beyond that, and that's where TSMC may be researching stuff they'll use a decade from now. It won't have any process name labels attached to it, and may not use standard fab equipment for certain steps depending on what they are doing. This is truly blue sky stuff where they can look at trends and say "well this transistor type is not going to work beyond another couple shrinks, what can we do next?" or "silicon is going to run out of gas, we need to experiment with different materials and see which ones work best", etc.
 

moinmoin

Golden Member
Jun 1, 2017
1,657
1,594
106
More to the point, if all of TSMC's competition fell away it would remove a lot of the incentives for them to keep pushing to try meet Moore's Law (or 1.7x of Moore's Law, at least) with a new process every two years.
The big difference with TSMC is that it is a pure play foundry, new nodes are its only product that makes its customers pay really big money. So even with less competition it's in TSMC's very own interest to keep pushing new nodes that have significant enough improvements for enough new customers willing to pay significant money for it.

As for the competition: Samsung handles its foundry like an expensive toy, GloFo floundered under the pressure AMD's demands and lack of size put on it, Intel was way too self-infatuated to notice that they were falling behind. For all of those foundries outside customers were of minor concern/influence leading to said developments.
 

LightningZ71

Senior member
Mar 10, 2017
417
314
106
Another thing that seems to work in TSMC's favor is their willingness to do what we tend to refer to as "half nodes". Intel likes to make big changes all across their process tech when they introduce a new node. It's a high risk/high reward strategy. 10nm is a raging dumpster fire for them because of bets they took on lithography tools and substrate materials (that we can tell from an outside view). Had they, instead, just elected to do a more modest node jump by just using a more refined set of tools with minor structure changes from 14nm down to something like an "11nm" node, then pushed for a materials change in a follow on node that we could call 9nm, maybe they wouldn't be in the shape that they are in. Their other issue is that they tend to deeply integrate the structure of their CPU dies with the characteristics of the node it was targeted for. This can allow them to extract more performance from that node, and processor, but its at the cost of risking the whole product if the node itself has a problem. This is one of the reasons that they are getting 5+ ghz out of their 14nm products.

On the other hand, TSMC is willing to develop and manufacture production lines for intermediate nodes that have only minor improvements from previous ones. This lowers the risk that they won't make any advances at all, and gives their customers a variety of process technologies to use for their products. Some nodes are better for products that need high durability, others are better for raw performance, others are better for low energy usage or maximum density. This keeps them in the game in many ways, and also makes them money on products that may not need the absolute best performance, but have other high end characteristics requirements. It also allows them to command top dollar for each new node as it is a clear advancement on previous ones and a more tailored fit for anyone interested in it. Node development is VERY VERY expensive. TSMC needs to command leading edge node prices to continue to develop additional new nodes. Chicken/egg.
 

Ajay

Diamond Member
Jan 8, 2001
7,064
2,364
136
Too much noise and not enough signal to figure out what’s really going on in Intel. Too bad Andy Grove couldn’t have been CEO forever. The only thing that makes sense to me is that too many Intel managers are more interested in their careers than they are in the success of the company; and management has failed to reverse that trend. I worked at a company that grew fast and then started to spend too much time on securing little fiefdoms rather than focusing on their products. It didn’t end well.
 

name99

Senior member
Sep 11, 2010
214
201
116
I think the MBA meme is more of a reflection of Kochs, Mellon Scaifes, etc spending the decades under the Great Compression (1940-1980) scheming to turn America into an oligopoly, complaining about how America had turned into a socialist paradise and it was stifling their freedom (to rule over others). Spending money influencing academics and funding think tanks to their cause to reverse public policy for their own gain, but ended up creating an economic dogma of short term maximization, and finally succeeded after packing the Supreme Court in the 70s with their victories in Belloti v Bank of Boston and Buckley v Valeo and then attaching Milton Friedman to Ronald Reagan's administration.

I think you'll find "MBAs" from before the 1980s weren't operating with this sort of "loot the business for my golden parachute" philosophy at all, because they hadn't been indoctrinated by philosophies funded by the extra money billionaires got from tax cuts yet.
If you want to turn this into a parable about the specifics of current politics, you'll miss most of what I am saying.
The greedy have always been with us, they are not new.
The point is not so much greed as cluelessness. The engineering mindset
- understands that the world is constantly changing
- understands that markets that appear small today will be larger tomorrow
- understands that you have to invest in R&D today to get a payoff tomorrow

The problem with these MBA managers is not that they are greedy, it is that they are incompetent. They are too stupid, too ignorant of their own industry, to understand the points I made above. So they make a few tens of millions for themselves, but don't make the billions that acquire to the engineer-entrepreneurs.
I am against STUPIDITY, rather less against greed.

And this is not a post 80s thing. In the 70s' Xerox showed the same blindness (office automation), and IBM showed the same blindness (mini-computer). It's perhaps more obvious now, but that's mainly because the world is changing faster now than in the 70s, and so the consequences of stupidity and ignorance of your own industry are visible faster.
 

name99

Senior member
Sep 11, 2010
214
201
116
SMIC sure as hell can't do it. Samsung is slowly retreating into also-ran territory. Intel just dropped out (effectively). GF? Please.
...
In the near-future, TSMC has been pretty candid and informative. I can get a lot more useful information on a node coming out in 6 months to a year from TSMC than I can from anyone else on anything else in the semiconductor industry.
SMIC can't do it? Oh yeah?

I'd bet a lot more on SMIC than on Intel...

Obviously SMIC are starting from way behind, and are still building up a complete pool of expertise across every aspect of running a modern fab and providing a foundry service. But they have the customers, they have the talent pool, they have government support and, for now at least, they appear to be run on engineering lines rather than short-term-profit-maximization lines. (Even though SMIC is a publicly listed company, right now it appears to be run more like TSMC than like Intel.)

As for TSMC and the future, you seem determined to pick a fight even when there is nothing to fight about.
Yes, of course TSMC talk about what they are doing six months to a year from now. What they don't talk about (and which Intel do all the time, and Samsung do increasingly) is what they plan to do further (sometimes much further) out than a year.
If you can't tell the difference between TSMC's bland few sentences on what they will be doing in a year and Intel's full court press advertising blitz of their roadmap for the next ten years (and more importantly, can't tell the difference between the mindsets that produced each), well, what can I say?
 

A///

Senior member
Feb 24, 2017
201
75
71
10nm is more than just late, it's still broken.
It's half-broken. They have some 10nm parts out, albeit mobile. This wouldn't be an issue if 10nm were based on a new µarch and they could lend a 40-60% IPC increase at those reported clocks which would obviously put it in line with a new architecture, and one we haven't seen in close to 20 years. The rest of your post is explaining stuff I know already or repeating stuff already said. I sold all my Intel shares nearly a year ago. I already knew how bad 7nm was.
 

naukkis

Senior member
Jun 5, 2002
344
171
116
It's half-broken. They have some 10nm parts out, albeit mobile. This wouldn't be an issue if 10nm were based on a new µarch and they could lend a 40-60% IPC increase
It would be still issue, problem is that Intel can't manufacture chips it needs with 10nm, and producing those chips it can is extremely expensive.
 

A///

Senior member
Feb 24, 2017
201
75
71
Who says 2nm isn't due until 2029? That's the next step after 3nm, which is scheduled for mass production in H2 2022 (i.e. exactly the timeline you'd expect for Apple to use it for the A16) If it takes TSMC 7 years to take the next process step after N3, Intel will have ample opportunity to catch up and retake the lead!
Yeah, my bad. I mispoke. I read the wrong section off of my notebook. Intel has 2-1.4nm in play in 2029 per that janky roadmap. I remember the TSMC press report from June.
 

Doug S

Senior member
Feb 8, 2020
250
301
96
The big difference with TSMC is that it is a pure play foundry, new nodes are its only product that makes its customers pay really big money. So even with less competition it's in TSMC's very own interest to keep pushing new nodes that have significant enough improvements for enough new customers willing to pay significant money for it.

As for the competition: Samsung handles its foundry like an expensive toy, GloFo floundered under the pressure AMD's demands and lack of size put on it, Intel was way too self-infatuated to notice that they were falling behind. For all of those foundries outside customers were of minor concern/influence leading to said developments.
If everyone else stepped off the train and TSMC stood alone and slowed down new process development to the point one came out a decade they could keep charging the same rates for the same process - or even increase them each year. What choice will their customers have when they need chips made at the best currently available node?

Regardless of whether better processes are available to them or not, Apple, Intel, AMD, etc. still need to make new phones/PCs/servers, and will need chips to put in them. If they aren't really much faster than the chips in last years models it may reduce demand for new phones and PCs but those devices still break/wear out or have other improvements not related to process technology that will induce people to buy new ones. Intel kept rehashing 14nm for about four years with very little performance improvement but they kept reporting record sales.

You don't understand why monopolies are bad if you think that TSMC, if they were a monopoly with the best process out there, couldn't charge high rates without a new process to justify it. That's how things work now, because there is competition. Without it, they have captive customers who need their chips, since they couldn't go anywhere else TSMC could charge whatever they wanted that would maximize their profit.

For example, let's say today they could sell 1 billion leading edge chips per year at an average price of $30/ea for $30 billion in revenue. If they had no competition they would set their price at the point where revenue is maximized. So maybe they try doubling it to $60, and find they have demand for 700 million chips (the other 300 million get made on older slower nodes or fewer phones/PCs/etc. are sold because they cost more) That's $42 billion in revenue which is a lot better than $30 billion. Let's try again with $90 and see demand for 500 million - that's $45 billion, even better. Try $120 and maybe demand falls to 350 million reducing revenue, so they end up settling in the $90 to $100 range. That's what monopolies do, and that's why there are laws against them.
 
  • Like
Reactions: Tlh97 and Saylick

DrMrLordX

Lifer
Apr 27, 2000
16,023
4,970
136
I'd bet a lot more on SMIC than on Intel...
SMIC will continue to advance as long as they steal appropriate technology from other parties.

As for TSMC and the future, you seem determined to pick a fight even when there is nothing to fight about.
Hmm, that was the impression I got from your post actually, but hey whatever.

Yes, of course TSMC talk about what they are doing six months to a year from now. What they don't talk about (and which Intel do all the time, and Samsung do increasingly) is what they plan to do further (sometimes much further) out than a year.
If what they're saying in the near-term is accurate, I'll take that over companies that have 5+ year roadmaps that aren't worth the pixels used to display them.
 
  • Like
Reactions: Tlh97

Zucker2k

Golden Member
Feb 15, 2006
1,158
514
136
Murthy bites the dust....

So, process development continues.
Edit:
 
Last edited:

moinmoin

Golden Member
Jun 1, 2017
1,657
1,594
106
If everyone else stepped off the train and TSMC stood alone and slowed down new process development to the point one came out a decade they could keep charging the same rates for the same process - or even increase them each year. What choice will their customers have when they need chips made at the best currently available node?
The choice is not to go to the bleeding money node. As the overall majority of customers already do currently. Apple as a first mover client would be instantly looking elsewhere if TSMC ever were to pull what you fantasize there. Everybody else would follow suit. End consumers wouldn't get the kind of node improvements seen in all the years before, but as you pointed out they didn't care much when Intel was doing that. TSMC can't pull the same since it relies on its clients to make use of its nodes for their products, if its clients feel they don't get their money worth at TSMC nobody is forcing them to stay there. TSMC can't turn around and fill their foundries with a product of its own making.
 
  • Like
Reactions: spursindonesia

DrMrLordX

Lifer
Apr 27, 2000
16,023
4,970
136
Whoa. Swan wiped out Murthy? Blood is in the streets.

@Kenmitch

That image . . . is that real? Or did someone photoshop that?

@alcoholbob

Interesting, which process, I wonder? N7? N7P? N6? N7+ and N5 are probably already committed. Also this still doesn't answer the question of what Intel is going to do with their own fabs.
 

ASK THE COMMUNITY