[EETimes] TSMC starting 5 nm EUV Risk Production in April?

[Ajay]

It would make a ton of sense if the public availability of Cooper Lake was a year from now, and 7 nm Granite Rapids two years from now. I don't know if that's realistic.

This roadmap (from AT) from last summer indicates that Ice-Lake and Cooper-Lake are both due in 2020, though Intel roadmaps have been confusing lately:

 

[Ajay]

https://fuse.wikichip.org/news/2207/tsmc-starts-5-nanometer-risk-production/

Wikichip estimates that TSMC's 5 nm node has a logic density of 171.3 Mtr/mm2; compared to 96.27 for their 7 nm. Intel's marketing had put their 10 nm node at 100.8.

Well, Intel 10nm is going fall shy of that number, and , from what I've read, Intel is planning on being less aggressive on density with 7nm EUV than they were at 10nm. That may not matter as much in the x86 CPU space, where xtor performance is also very important.

 

[jpiniero]

This roadmap (from AT) from last summer indicates that Ice-Lake and Cooper-Lake are both due in 2020, though Intel roadmaps have been confusing lately:

Intel's roadmap would make more sense if you realize the 10 nm products aren't going to be in any kind of volume compared to the 14 nm products. Icelake Server especially is going to be nothing more than a distraction, corps if they aren't buying Rome will be buying Cooper Lake.

 

[Ajay]

Intel's roadmap would make more sense if you realize the 10 nm products aren't going to be in any kind of volume compared to the 14 nm products. Icelake Server especially is going to be nothing more than a distraction, corps if they aren't buying Rome will be buying Cooper Lake.

I guess I still have problems with that, as in, why is Intel even bothering with 10nm at all. If it’s that bad? Why the farce? Why the waste of resources for so little return?

 

[jpiniero]

I guess I still have problems with that, as in, why is Intel even bothering with 10nm at all. If it’s that bad? Why the farce? Why the waste of resources for so little return?

They might have had some contracts to fulfill, but my guess is mostly Shareholder PR. Thing is, the resource usage shouldn't that much, they are converting most of the 10 nm production lines that had been setup back to 14 or repurposing it for EUV.

The scary thing for Intel that it's not out of the question that AMD could have 5 nm server parts out and they could still be peddling 14 nm products more or less.

 

[moinmoin]

I guess I still have problems with that, as in, why is Intel even bothering with 10nm at all. If it’s that bad? Why the farce? Why the waste of resources for so little return?

Presenting nothing new at all due to canceling everything 10nm would be a very bad look, so they do the opposite extreme and are drowning everybody in a whole slew of new product announcements which, if one looks closer, still amount to essentially nothing new at all. As long as the stock markets play along this will allow Intel to sufficiently bridge the gap until 7nm is up and running.

 

[Ajay]

The scary thing for Intel that it's not out of the question that AMD could have 5 nm server parts out and they could still be peddling 14 nm products more or less.

If AMD sticks to a two year cadence on servers, then Intel has a chance to intercept with 7nm CPUs (and EMIB). It’s weird, wondering if Intel will get 7nm done in time - things sure have changed.

 

[Ajay]

Presenting nothing new at all due to canceling everything 10nm would be a very bad look, so they do the opposite extreme and are drowning everybody in a whole slew of new product announcements which, if one looks closer, still amount to essentially nothing new at all. As long as the stock markets play along this will allow Intel to sufficiently bridge the gap until 7nm is up and running.

That makes sense, if investors buy the 'story', Intel doesn’t face a massive shakeup and can keep pursuing their current plans. Otherwise, the sh*t hits the fan.

 

[moinmoin]

That makes sense, if investors buy the 'story', Intel doesn’t face a massive shakeup and can keep pursuing their current plans. Otherwise, the sh*t hits the fan.

So far not only investors but the whole market buys the ongoing story. Just imagine how Intel's range of products would look like today if they weren't doing any of the continuous refreshes of refreshes on 14nm, all of which would not exist were 10nm realized on time as in the initial plans. Now TSMC caught up, allowing for unprecedented competition, while EUV is potentially delayed yet again (an issue for both Intel and Samsung since they skipped the non-EUV 7nm step).

 

[Ajay]

Now TSMC caught up, allowing for unprecedented competition, while EUV is potentially delayed yet again (an issue for both Intel and Samsung since they skipped the non-EUV 7nm step).

My google-fu has failed me, what is this potential delay in EUV?

 

[moinmoin]

My google-fu has failed me, what is this potential delay in EUV?

Nothing concrete, it's just a fact that EUV still hasn't been used to mass produce a whole chip by anybody yet. Apple supposedly started mass production on A13 on TSMC's N7+ (7nm with EUV only used for up to four uncritical layers, mainly transistors, not yet interconnects) in March. TSMC's first complete EUV node will be called N5 planned to start mass production next year (again with Apple as primary customer, for its A14 chip). Samsung started risk production on its EUV based 7LPP late last year, no news since, they may actually be first with full EUV chips depending on progress this year. Intel's 7nm is even more up in the air so far. Regarding supply with EUV tools that potentially affect production, an ASML supplier suffered a fire late last year.

 

[jpiniero]

Apple isn't using 7+ for the 2019 iPhone, it's just the regular TSMC 7.

 

[DrMrLordX]

Apple isn't using 7+ for the 2019 iPhone, it's just the regular TSMC 7.

Is the 2019 iPhone using A13?

 

[jpiniero]

Is the 2019 iPhone using A13?

I would think so. It's just not using 7+, just regular 7.

 

[DrMrLordX]

I would think so. It's just not using 7+, just regular 7.

Hmm. Macrumors seems to think A13 is still 7nm+:

https://www.macrumors.com/roundup/2019-iphones/

A13 Processor

TSMC will be the sole supplier of the next-generation "A13" ships that will be used in the 2019 iPhone lineup, and iPhones coming in the future. The A13 is expected to be built on TSMC's 7nm+ package with extreme ultraviolet lithography.
The A13, like all chip upgrades, should bring performance and efficiency improvements.

PocketNow seems to think TSMC will begin shipping 7nm+ product in H2 2019:

https://pocketnow.com/7nm-chip-orders-ramping-up

Mass shipments of EUV-based 7nm chips are expected to kick off in the second half of the year as well.

Dunno when the next iPhone products will ship, but the TheInquirer seems to think September:

https://www.theinquirer.net/inquire...ecs-price-chassis-leak-foxconn-triple-cameras

 

[jpiniero]

Oh they are wrong. TSMC said in an earnings conference call they don't expect any meaningful revenue for 7+ in 2019; not until 2020.

 

[moinmoin]

Oh they are wrong. TSMC said in an earnings conference call they don't expect any meaningful revenue for 7+ in 2019; not until 2020.

Thanks. So I guess there will be no EUV products at all (? pending the meaning of "meaningful revenue") in 2019, unless Samsung surprises us.

 

[Ajay]

If this is true, then surely 5nm EUV for TSMC is pushed back as well. Seems kind of strange to announce risk production for a lithography process that is now likely two years out (unless their big foundry customers are all skipping 7nm+.

 

[jpiniero]

If this is true, then surely 5nm EUV for TSMC is pushed back as well. Seems kind of strange to announce risk production for a lithography process that is now likely two years out (unless their big foundry customers are all skipping 7nm+.

The two are coming at the same time basically. Part of that is because EUV is the bottleneck and not the node design. So it appears that 7+ is more for customers who don't want (or can't) pay the crazy design costs of 5. The big customers like Apple will skip 7+.

 

[DrMrLordX]

Oh they are wrong. TSMC said in an earnings conference call they don't expect any meaningful revenue for 7+ in 2019; not until 2020.

How long ago was that? It seems like major publications didn't pick up on that story (and shame on them for not doing so).

 

[jpiniero]

How long ago was that? It seems like major publications didn't pick up on that story (and shame on them for not doing so).

Don't remember but it was pretty recent. From what I remember, the TSMC exec said they will do decent business with 7+, just not in 2019.

 

[trivik12]

https://seekingalpha.com/article/42...wei-q4-2018-results-earnings-call?part=single

 

[dark zero]

A13 2019 h2 on 7+ then? A14 2020 h2 on 5nm?
Amd x86 half a year later than Apple.
Tock tock tock tock. Relentlessly.
It's perfectly clear Intel business model is not working today. In no way can they compete with the rest of the world sharing process cost.
They need to change something radical now or they will be stuck on a half mature complex 10nm while the rest of the leader pack is on 5nm euv. It's not viable.

And making it worse, EUV will be as good as the Intel chips on that manufacturing dimensions (5nm). EUV is the real deal in fact.

7 nm EUV would be similar to 7 nm Intel chip.

Intel actually is on real trouble. They are great but they are also slow.

 

[Ajay]

And making it worse, EUV will be as good as the Intel chips on that manufacturing dimensions (5nm). EUV is the real deal in fact.

7 nm EUV would be similar to 7 nm Intel chip.T

Intel actually is on real trouble. They are great but they are also slow.

How do you know this? Is there data on dimensions and performance for Intel 7nm and TSMC 5nm??

 

[krumme]

Intel is reliant on extreme margins and profit from data centre group to fund process development. If not they are in a very bad position. Imo it's a very risky position because its vulnerable.

If zen 2 (edit from zen3) get latency under control and manages a more consistent pattern of performance that zen1 lacks for many server loads where does that leave margins...profit ...and thereby future funding of process?

Every one and their brother minus a few will continue to fund apple because its apple but you simply can't sell a server because its Intel. It's different markets. Consumer vs. Pros.

The business environment is also far different today than 15 years ago. More specialised. More professionalized. Cut throat focus on cost vs perf. Everyone gets fired for buying too expensive no matter the name.

Intel can muster up the best saleaforce and technical marketing and support but it's not 2004 any more. They need superior solutions and the bare silicon is just key today. They flat out need a superior cpu arch. End of story. Cut the rest.

The market is more red. It was bound to happen as all markets mature but I think that will be more of a wakeup than zen3 perf in itself. Zen3 might just trigger or emphasize that change in the market. For some reason those changes in markets always come as a surprise to those involved.

The zen3 server bm - especially latency and performance consistency - can be a bigger drama than we imagine.