10nm yield problems

[lefty]

Rumors are that all foundries (Intel, TSMC and Samsung) are having problems with their 10nm node.
Yield problems at TSMC has caused the delay of MediaTek's Helio X30. Also, Yield problems at Samsung have pushed back the launch of the Galaxy S8 smartphone. This is according to Digitimes: Yields of 10nm FinFET Chipsets Are Still Low – SoC Suppliers Forced to Reschedule Shipment Dates (sorry, link has been paywalled, this is link to WCCFTech instead)
Meanwhile, is seems possible that Intel faces yet another delay on their 10nm node. BlueFin Research Partners have heard that "INTC is still struggling with defect/yield issues with this advanced process for High Volume Manufacturing (HVM) release and have not locked down the complete process of records (PORs) to date".

 

[astilo]

I really wonder if the yield issue is mainly defect density driven or more related to some device variability issues. I mean, as far as I know, foundry defect density at 14nm was pretty good, so at 10nm although not great, it should not really be a show stopper. On the other hand, quadruple patterning should not really be a trivial exercise (overlay specs must be a nightmare).

 

[Daniel Nenni]

I don't believe the 10nm bad yield rumors. Last week I spoke with the equipment people at SPIE and the EDA folks at Dvcon. Next week will be the true test at the TSMC Symposium with all of the IP people. IP doesn't lie.

FYI, TSMC is not using quad patterning for 10nm...

 

[astilo]

[FONT=&quot]MediaTek just rolled out its 10nm Helio X30 SoC-series. The company expects smartphones using the chips to be available in the second quarter of 2017.[/FONT]
[FONT=&quot]MediaTek co-COO Jeffrey Ju was quoted in a recent [/FONT]Nikkei [FONT=&quot]report as saying the Helio X30 chip is "a bit delayed because the yield rate of 10nm process technology has not yet reached a satisfactory level."[/FONT]

 

[Fred Chen]

The same rumor mill now says MediaTek is moving on to 7nm regardless: MediaTek sticking with TSMC for 7nm chips

MediaTek sticking with TSMC for 7nm chips
Cage Chao, Taipei; Jessie Shen, DIGITIMES [Thursday 9 March 2017]

Despite experiencing unsatisfactory yield rates for TSMC's 10nm process, MediaTek will continue to partner with the foundry to make its next-generation flagship mobile chips, according to industry sources.

TSMC will enter risk production for a MediaTek chip on the newer 7nm process in the second quarter, said the sources, adding that the SoC incorporates 12 CPU cores.

TSMC disclosed previously the node technology will be ready for risk production later in the first quarter.

In addition, speculation has circulated that the low yield rates for TSMC's 10nm process could affect shipments of MediaTek's Helio X30 chips. In response, MediaTek claimed there is no change to the shipment schedule.

MediaTek in February rolled out its 10nm Helio X30-series SoC designed for high-end smartphones. Mass production for the deca-core SoC has kicked off, said the company, adding that commercial devices powered by the chips will be available in the second quarter of 2017.

 

[maks10]

yes, MTK has started some activity with TSMC on 7nm.

 

[Fred Chen]

On March 29, Samsung is set to unveil the Galaxy S8 (Samsung Confirms Galaxy S8 Announcement For March 29 | Ubergizmo). The Galaxy S8 has been expected to use either the Snapdragon 835 or Exynos 8895. It is expected to go on sale April 21, though initially there was a reported week's delay to April 28 (KGI corroborates Galaxy S8 specs & April 21 release date, projects slower sales compared to Galaxy S7 | 9to5Google).

At MWC 2017, Snapdragon 835 processor was quite absent, so many phones showcased 14nm Snapdragon 821 instead (MWC 2017: How the Snapdragon 835's Absence Stole the Show). This fueled the 10nm yield rumors.

If the Galaxy S8 release is without 10nm or otherwise delayed, that would indeed confirm the rumors.

 

[Portland]

Cannonlake will be released second half of 2017.

 

[Daniel Nenni]

I really wonder if the yield issue is mainly defect density driven or more related to some device variability issues. I mean, as far as I know, foundry defect density at 14nm was pretty good, so at 10nm although not great, it should not really be a show stopper. On the other hand, quadruple patterning should not really be a trivial exercise (overlay specs must be a nightmare).

TSMC stated that 10nm is now in HVM meaning no yield problems. Mediatek may have gotten ahead of themselves. 7nm is also on track for early 2017 HVM, just in time for Apple iProducts of course.

 

[dennys822]

On March 29, Samsung is set to unveil the Galaxy S8 (Samsung Confirms Galaxy S8 Announcement For March 29 | Ubergizmo). The Galaxy S8 has been expected to use either the Snapdragon 835 or Exynos 8895. It is expected to go on sale April 21, though initially there was a reported week's delay to April 28 (KGI corroborates Galaxy S8 specs & April 21 release date, projects slower sales compared to Galaxy S7 | 9to5Google).

At MWC 2017, Snapdragon 835 processor was quite absent, so many phones showcased 14nm Snapdragon 821 instead (MWC 2017: How the Snapdragon 835's Absence Stole the Show). This fueled the 10nm yield rumors.

If the Galaxy S8 release is without 10nm or otherwise delayed, that would indeed confirm the rumors.

The analyst, Andrew Lu had comment on his FB yesterday. But seems he changed his word compare to his initial one. In previous, looks like pretty sure that SEC is ahead TSMC in 10nm.

陸行之 ～ Andrew Lu on global semis and techs | Facebook

 

[lefty]

This is the full quote from Andrew Lu's facebook page:

Does Samsung Lead TSMC 10nm Ramp by 3 Months and Cut TSMC’s Mid-Node Advantages on 8/​6nm Introductions?
Last week, Samsung announced to have shipped more than 70,000 wafers of its first-generation 10nm LPE (Low Power Early) to date with a steady high yield to meet customer needs. Samsung Electronics has also announced the addition of the 8nm and the 6nm process technologies to its current process roadmap of 10nm and 7nm.
If Samsung is telling us the truth, Samsung is leading ahead of TSMC on 10nm ramp for a quarter (vs. Samsung's lead on 14nm ramp for six months). Thus, the next check point is if Samsung is able to ship its Galaxy S8 globally before end of May 2017.
Also, different from most of the foundries, TSMC often introduced its mid-node technology like 12nm in between of 10/16nm and 22nm in between of 20/28nm to gain its competitiveness on performance and cost. However, if Samsung is also introducing its mid-node like 8nm in between of 7/10nm and 6nm in between of 5/7nm to compete with TSMC's mid-node, this leaves little room for TSMC to gain any technology edge.

He hasn't heard about TSMC's 7nm+ then. That would be equivalent to Samsung's 6nm.
Samsung have 8nm, but they don't have 12nm, or 22nm - so overall not very clear that they are ahead.
(BTW. 12nm is not "between 10/16nm", it's a 6T version of 16nm and 22nm is not "between 20 and 28nm". 22nm is a improved version of 28nm)

 

[Daniel Nenni]

This is the full quote from Andrew Lu's facebook page:

He hasn't heard about TSMC's 7nm+ then. That would be equivalent to Samsung's 6nm.
Samsung have 8nm, but they don't have 12nm, or 22nm - so overall not very clear that they are ahead.
(BTW. 12nm is not "between 10/16nm", it's a 6T version of 16nm and 22nm is not "between 20 and 28nm". 22nm is a improved version of 28nm)

22nm is an optical shrink of 28nm. Just like 55nm is an optical shrink from 65nm. This really is a clever marketing move by TSMC. 12nm is not an optical shrink so work will have to be done to get a design moved over.

I read the Andrew Lu piece and found it to be complete nonsense. Tom Dillinger will be posting his Top Ten from the TSMC Symposium so look for that this week. There were some definite nuggets that the mainstream media missed.

D.A.N.

 

[lefty]

22nm is an optical shrink of 28nm. Just like 55nm is an optical shrink from 65nm.

That's good to know.
One thing I forgot, is that Samsung is meant to be using gate all around transistors for 7nm. That is something that could give them an edge

 

[Staf_Verhaegen]

This really is a clever marketing move by TSMC.

I don't like it, it makes my engineering heart bleed. Also some double standards seem to be at play here on SemiWiki; sometimes it's called 'alternative facts' and sometimes 'clever'.

 

[simoncc2]

22nm is an optical shrink of 28nm. Just like 55nm is an optical shrink from 65nm. This really is a clever marketing move by TSMC. 12nm is not an optical shrink so work will have to be done to get a design moved over.

I read the Andrew Lu piece and found it to be complete nonsense. Tom Dillinger will be posting his Top Ten from the TSMC Symposium so look for that this week. There were some definite nuggets that the mainstream media missed.

D.A.N.

28nm is already an optical shrink of 32nm ... is that the first time in the industry that a node is shrunk twice ?

 

[Daniel Nenni]

28nm is already an optical shrink of 32nm ... is that the first time in the industry that a node is shrunk twice ?

28nm is not a shrink, it is HKMG. TSMC did not even do a 32nm, they went from 40nm to 28nm HKMG. Maybe they will shrink 40nm to 32m ULP for IoT but probably not.

 

[Fred Chen]

On March 29, Samsung is set to unveil the Galaxy S8 (Samsung Confirms Galaxy S8 Announcement For March 29 | Ubergizmo). The Galaxy S8 has been expected to use either the Snapdragon 835 or Exynos 8895. It is expected to go on sale April 21, though initially there was a reported week's delay to April 28 (KGI corroborates Galaxy S8 specs & April 21 release date, projects slower sales compared to Galaxy S7 | 9to5Google).

At MWC 2017, Snapdragon 835 processor was quite absent, so many phones showcased 14nm Snapdragon 821 instead (MWC 2017: How the Snapdragon 835's Absence Stole the Show). This fueled the 10nm yield rumors.

If the Galaxy S8 release is without 10nm or otherwise delayed, that would indeed confirm the rumors.

So it has been confirmed by Samsung at their release event on March 29 that Galaxy S8 will​ contain 10nm and be available April 21 (presumably in US). The next to watch will be how soon these run out...

 

[IanD]

28nm is not a shrink, it is HKMG. TSMC did not even do a 32nm, they went from 40nm to 28nm HKMG. Maybe they will shrink 40nm to 32m ULP for IoT but probably not.

If TSMC 28nm isn't a shrink, why is there a x0.9 scaling factor from drawn to silicon dimensions, as anyone who's done layout in it will be well aware?

IIRC TSMC originally intended to have a 32nm process and that's how the rules were designed, then decided to do a 10% shrink to 28nm get better performance/density, then the 32nm process disappeared because everybody wanted 28nm, but redefining all the design rules which everyone was using would have been too much effort.

 

[simoncc2]

So it has been confirmed by Samsung at their release event on March 29 that Galaxy S8 will​ contain 10nm and be available April 21 (presumably in US). The next to watch will be how soon these run out...

seems that the new Ipad will be released before the S8 ( around 10th of April)with 10nm processor. so TSMC is also in good shape. However the volume are lower ...

Pretty interesting to see at the end that the first consumer products in 10nm in TSMC and SEC will be on the shelf within few weeks I'm sure Daniel would say it's the Fabless industry success, "Absolutely"